To get things started, let's look at examples of decision support in particular. As I go through these examples, I want you to think about who has the decision? What is the decision? Whether the screen you're looking at matches the decision to be made? Let's start off with something you know. You've seen a popup like this all the time. You've deleted the file, The machine ask whether or not you want to continue. So, this is actually decision support. It's a popup alert. It's what is called hard stop. You can't continue unless you actually respond to it. The machine is giving you three different responses on the same popup. Who has the decision? That's the user. What's the decision? Whether you want to retain the information that you say you closing or not. I would say that this interface matches the decision to be made. The hard stop seems reasonable because if you ignore it, you may lose data that you really don't want to lose, and you'll be really upset. So, I think the cost of the pop-up is worth the gain, and it is directed to just the user, and it gives you the options that you should need. So, it's nice and clean as opposed to the interfaces that you will see with a lot of other health decisions support. So, let's look at some other decisions support. So, this is a poster on the wall of clinics dealing with diarrhea in countries where there's a lot of diarrhea. The first thing to get across here is that this is not a computer. So, a poster is as much as an intervention, and it can be as much as decision support as anything you can get on the computer, so the principles should be the same whether you're talking about a piece of paper, or tag, or a fancy computer system. So, the idea here is that the patient comes in, gets evaluated, and depending on the evaluation, gets triaged into either a green door, don't worry about it, yellow's worry about it, and red, worry about it a lot. So, who's got the decision? I think it's clear that it's the clinical staff. I don't know if it's a doctor or not. It doesn't matter. It's written in such way that somebody with relatively minimum training could apply it. The decision to be made is the triage which literally means decision among three options. In terms of matching, I think it's pretty clear. The folk is very clear, the data you need to collect is very clear, and the implications of each decision is color coded and clear. So, I think this is a great example of decision support even if it's out of a computer. But we want to get to computers, so here's one. This is from an app that is made available to all electronic health records through something called Fire which we will be talking in more detail later on in the course. So, what you can see is a patient just coming in with chest pain and this screen displays data that's relevant to the decision around chest pain. Now, you can see that there's not a lot of data being posted. There is some information about previous EKGs or echocardiograms, and then you see the tracing on the recent EKG on the screen. So, who's got this decision? It's some, again, a triage, this is not for diagnosing chest pain. This is basic to see. If something horrible goes on right now, which way should we go? I would say it's a physician because they're the ones who usually know how to read an EKG, those tracings. Most nurses do not, but maybe a particular nurse could be trained in that way. So, the decision to be made is the triage decision. As I said, this is too little detail for diagnosis, and is there a match? Well, I would argue that a little bit depends on a locale you're in. Some locales would have more laboratory data may be available. It might depend on how far in the triage decision is made by the staff. Is it upon coming in? Is it allowed to be an hour later? It's not clear what the match is between who's got the decision, the workflow that this site has for managing chest pain. I would argue as well that it is missing a lot of data even to manage just the triage decision not alone the diagnosis decision. This is an example of what's sometimes called a nomogram or a curve. So, let me give you the story here. Newborn babies can accumulate pigment in their blood called bilirubin. If that pigment builds up too much, it goes into the brain, stains the brain yellow or green and causes irreparable brain damage. So, that's called kernicterus. Icterus is jaundice. Kern is the center of your head, your brain. So, kernicterus means yellow brain, not something you want your child to have. We spend, in pediatrics, a lot of effort to preventing kernicterus in particular and you can see that the levels of bilirubin where you really have to do something drastic immediately is in the red area on the top of this screen, on the top of the graph where the x-axis is age since birth and the y-axis is level of bilirubin. You can see that the actionable areas change as the baby gets older. So, even though the top levels were absolutely, the lower levels do change. The lower levels, what do you do before the red area? Let's be clear. The red area is called an exchange transfusion. We had to put an IV through the belly button, and you have to put in five cc's of good blood, take out five cc's of the bad blood, and go back and forth for it's called double volume exchange. It's a big deal. You don't want to have to do that. So, it turns out that a bunch of nuns back in the 1950s realized that when babies were in sunlight, their level of kernicterus was lower. They didn't develop as bad hyperbilirubinemea, high bilirubin in the blood, and so there are these things called bili lights so you can put a baby under, and that's what you do in the orange and maybe even yellow areas. So, the idea here is you plot the baby's blood levels over time and you follow the recommendations based on the color. So, what is the decision? So, here's a picture of bili lights. There is a baby just hanging out under the blue lights. So, the decision-maker is generally the pediatrician. Is it the attending of a senior physician? Is it a resident? Is it an intern? That depends on the locale. The decision is clearly whether to put in bili lights or to do transfusion or to do nothing. Not all decisions are triple decisions like this, but I guess it turns out this way. Is there a match? A little bit depends on the locale because maybe some places the bili lights are not that available, so they may be willing to tolerate higher levels in other places. Pretty other some subtleties. I should point out that this graph was designed for healthy newborns without risk factors for the kernicterus. So, it says it on the graph, but the user has to recognize that there's some babies to prove this graph just does not apply. Another example. We've seen this before. You're ordering drugs, the patient is allergic to drugs, and you get a popup, patient is allergic to drugs. What do you want to do? This is not much different, is it? From, you're about to delete your files. Do you want me to delete the file or not? You can see in this interface, we've given a lot of choices about why you might want to go ahead with the alert or not and then you can click, "OK". So, who's got to decision? I think the ordering clinician. What is the decision? Whether to change it dose, change the drug, or stick with the regiment I guess another triple, a triple decision. Sorry about that. Is there a match? Well, with a lot of these, very often it depends on what does clicking "OK" means. You might remember the case of the file with three buttons. One was save, the other was cancel, the other is can go ahead and delete. The triple decision and the buttons were there to support the triple decision. Here we have one button. It's not clear what you do, but happens if you click on that button. Getting closer to the patient, this is another one of these apps that's supposed to live on top of other environments. This is managing blood glucose in somebody with diabetes, and at the top you can see many days of blood glucoses, on the bottom you can see just one day or a finite time period and on the right hand side, this is a summary of everything. Again, you see red for bad, green for good, yellow for that's so great. Another point is that even the diabetes is a disease with high blood sugar, it's actually if you treat it and you go too low, that's what can kill you immediately as opposed to diabetes which unfortunately can kill you after a long period of time. So, what is the decision, who's got the decision in this interface matching? Well, we could start off with the managing clinician, right. The patient is bringing their blood glucose levels to a patient, and the question is whether or not to change the regimen for that triple decision. Is there a match? Well, on this graph, you can't see what the regimen is, you can't see what the injections are, you can't see what the patient was eating. So, we don't know whether the regimen is a good match for this patient or not. But I bet for the patient side of things, this patient is given these two circles, green clearly means a proportion of time you're in good area, yellow is the amount of time you are in the [inaudible] and red is danger. So, I think from a patient perspective, this is very helpful for helping convince them that the regimen might need to be changed because they can say, look at it you don't get a lot of green, we need to change things. It's not clear whether that how amount percent of red you're looking at is that too much, is that acceptable. Those levels of acceptability are not indicated on the graph or the graphic, so the patient may not be as motivated as you may want them to be. So, this kind of decision in this case we're going to say it's the patient, and the decisions whether to go along with the change that the physician is recommending and as I would argue it's not clear whether there's enough information to change the patient's mind. Just to show you that there are other ways of doing this, and this is from a study by MI Kim when he was here at Hopkins, he then went to Boston to continue practicing endocrinology. He devises interface which I think is one of the best ways of displaying patient data for a patient decision-making you've ever seen. It's worth spending a little bit of time on this. So, this is again a collapse of information over times rather than seeing the graph or the graph of blood glucose is over timing, showing reds and yellows and greens. Dr. Kim here has collapsed the blood sugars and give them this distributions. In other words, what proportion of the time is applied above normal, below normal and normal. But wait, there's more because these distributions are not just overall, but in fact they're yield to the regimen at the treatment that led to that distribution. So, at the top, you can see the lunch, he said wait a minute, why am I seeing lunch, should I be seeing breakfast first. Well, the insulin that the patient takes in the morning affects both the breakfast blood sugar in, while the after breakfast blood sugar in the pre-lunch blood sugar. So, when you look at the lunch distribution, you mean looking at the morning regimen. The breakfast regimen is actually a function of the previous slides, long dose insulin, et cetera. So, there's a lot packed into the slide and Dr. Kim showed that with interface like this, the treating physician changed the regimen to the optimal regimen more correctly and more quickly than these other distributional displays. So, the story here is that, the display really needs to match the decision that you need to make and requires deep thinking about what the problem really is, and what needs to be done. This is a population wide graphic of patients who need a vaccination or past due in red. So, I think you could see that there's quite a lot of patients who need the vaccinations. But let's think about this, who's got the decision here? Well, the era of population health very often is a care manager separately from the physician. It could be a physician is looking over a population, but very often it's the care manager who is dealing with the day-to-day issues of population health and he or she would be looking over this distribution and their decision is which patients should they tackle, just say focus on. Is there a match between the interface and the decision? Well, if in fact they can click on the red bars and see which patients will get a list or which patients that red bar corresponds with, then yes. If it's just a display of somebody who people are missing and then you have to go somewhere else, that's so enthusiastic. Then finally, to public health decision making. This is a dashboard if you will, assembled actually annually, and in this case they're looking at New York City where diabetes is the highest. You can see that map on the left and the right hand side, you are getting some quantitative assessment of how different these different zip codes are. So, I ask you, who's a decision-maker or master of decision? I think we will agree, is that the epidemiologists? Epidemiologists brings the evidence to the public health officer, it's a public health officer who makes the decision where to focus some amount of resources on. So, they have a pot of money for diabetes, this helps them with where to apply those resources, I think the biggest bank for the buck, whether is going to make the biggest difference. If they're looking at diabetes versus other opportunities or other needs in the city, than this particular dashboard would not help them with that decision. So, I think you got a good sense from looking at these that through the machine to help with decision-making, you want to know who you're talking about, you want to know what decision you really have to make, you see what data you're displaying and what affordance you're supplying. What is it that they can do? What's that I'm looking at the decision support is really important. So, the goals of this course then, are to help you help the user, whoever that user is make better decisions through better data, that better data displays a better affordance. That includes for the clinical decision making, share decision-making with the patient, in the case of administration, there's management decisions, in case of public health, public health decision making, and you'll see the same principles playing out. So, these principles are given to us by the AHRQ and pretty much accepted by the clinical decision making community, and I think they apply to the other communities as well. You want to get the right information to the right person in the right context. That could be work flow, could be the system, it can be time through the right format, what actually seen on the screen and again at the right point in the workflow. I said workflow twice, it's really important. Just for fun, that connects these to the stack, the five rights have to be reordered a little bit, but you can see who relates to the roles and perspectives, the when is the workflow, the where it involves the information system, the how is the extra module being used and what is being displayed involves data information and knowledge. So, it's good to know about the stack, so you can think about decision support in a coherent way and not just focus on one part. That's what we'll be worrying about the course of the course. So in summary, you will be called upon at some point to comment on something like decision support, to suggest it, to specify it, to manage it, to choose it, to build it, to buy it. So, we want to give you in this course some basic concepts we're thinking about and also to help you think about how you should do it. So, the outline of this course are the first four bullets will go into be tackling today. Should you build or buy it? Doesn't work. Once you've selected it, is it being used once you have it in place, and is it accomplishing your goals? Once we think about those, all that stuff which is really the top part of the stack, then we'll go back in and say, okay, now what is that we're talking about and how do you build it in particular. What's this business with knowledge and how do you get knowledge into these systems, and how do you get knowledge projected to the user.
