So, that's all about the rule itself. Then there are other rules that are connected to the main rule. If you recall, our little rule concluded that then patient is at critical risk. Well, what the heck does that mean? What do you do with that notion of critical risks? So, now you need a second rule that is consequent on the first, that then says, "Okay if the patient is at critical risk then recommend exchange transfusion." Now, different institutions could have different recommendations. One might say, "Recommend transfer to an institution that can do exchange transfusions." Another might say, "Recommend consult with a certain type of neonatologists or rheumatologists or whatever." So, what you do with that notion of critical risk is really a local issue. So, by having this consequent rule, I've divided generalizable knowledge like above 20 is critical risk which the world agrees on. From I've separated that knowledge for the more local business rule type of knowledge which is, if we're in critical risk this is what we do here. The first type of rule is more transportable. The second type of rule is less transportable. Both of them are vital for your decision support. But you can see why you might want the degree of freedom to separate the biological clinical knowledge from the Local Business School logic. By having the rules be explicit in this way, we're able to do that sort of separation that we can't do when you have everything buried inside code. But of course, rules are more complicated than a simple if A then B. So, here if you go to the orange part of the graph, we see things get a little bit more complicated. So, here I've drawn a rectangle that goes from zero to 12 hours and then it goes from like seven milligrams per deciliter up to 20. You can see that to fulfill that rectangle I can create a rule now if it transcutaneous bilirubin is above seven but below 20 and your age is less than 12 hours then you are high risk. So, you can see that we can make the rules more complicated and we can have a different conclusion and therefore different consequent rule exits. What is high-risk mean? What do you do? We're not going to go into that right now. Here's a second tape of a rule, an alert rule you may recall from our widget lists that I claimed that alert rules are under the list of using explicit knowledge. So, here you go. I could write the rule in this way that if the patient is allergic to penicillin and the physician has ordered penicillin then alert patient is allergic to penicillin. Now in this case, the alert sends, I could divide this rule into two but it seems pretty straightforward to simply have the alert. I put penicillin in italics because just like with transcutaneous bilirubin how does the machine represent this notion of penicillin? We'll be getting to that soon. Then I have this allergic to penicillin in not italics because it's a concept of allergic to penicillin. So, now I need to figure out before I separated the then statement into a subsequent rule. Here I need to separate the antecedent that left-hand side into two rules. How does the machine conclude that you're allergic to penicillin? That is more "business really" than the alert, because it all depends on where allergy is encoded in your system. It has to do with how much your system trusts the allergy field. It may depend on whether your allergy field is able to include level of allergy or level of belief for the allergy. So, there could be a lot of business rules around the concept of allergic to penicillin. So, I've separated out that rule. So, you can see that there's this notion of separating biological and clinical knowledge from what happens at your institution. Also, on the list of widgets was this notion of guidelines and protocols. I'm going to make the claim here that a protocol or a guidelines can be implemented or represented as a set of rules. So, if you look here, the context here is somebody comes in with a lump on their neck, good or bad, what I do with it. So, this is diagnostic decision support. So, you can see that this protocol starts off with, is it localized to the neck or not. If it is not then you ask about recent onset and then if it is you make a conclusion, if it's not you make another conclusion. So, in a sense, a pathway through this guideline or this tree is a rule. So, if it is not localized but is recent then conclude. If it is if it is localized and is not recent then conclude something else. So, there are four pathways in this particular classification tree and therefore there are four rules. Now, this looks like a flowchart in which case the if-then boxes should be diamonds. But it's a little bit different because it's really about it's classification. But the main point I want to get across here is that there are other ways of representing knowledge but to make it actionable to the computer generally you want to turn it into rules. I've been talking so far about biology and clinical knowledge and local business knowledge. Clearly, this is an example of clinical biological knowledge. This is published. These are what the best experts say. What you do for let's say post viral lymphadenopathy. One institution may do something different than other institution what you do, but CMV again could be local experience, local knowledge, local expertise. So, again we're separating out local from generalized knowledge and again having explicit rules makes it easier even though it's the same format. So, if you look at some of the other widgets. So the task assistance, while one can build them with this screen one could also, create them as rules. Many of these are implemented as flowcharts. So, I think I have convince you that I can make a rule for a path for the flowchart, reactive alert we show you already, diagnostic suggestions, I hinted that already just with this [inaudible] example. A performance dashboard is a different type of user that a clinical users as much as some sort of management user. But if this key indicator or some rate is above or below threshold and you got to take some action. So, it could be that if the rate is below a threshold than alert, what does alert mean? Does alert mean you send somebody to the household to slaughter the favorite horse and put that horse's head in a bed like in "The Godfather" or do you send a note to the department chair or do you send a little quiet no to the physician? There are many ways that one can alert. But certainly, the manager wants to know at least one will be above or below a threshold. Now, what the threshold is could be something that comes from the outside or it can come from the inside and there's actually a decision itself of the institution. Order facilitators and these guidelines again I showed these to you as screens that can be architected. But they could also, have rules behind them that are explicit, because then if the guideline changes, it makes it easier for the analyst to say, "Oh this consequent needs to be removed from the rule without having to manipulate screens which is an auditable." Similarly, for data displays and documentation forms, there could be rules behind all of these. Now, when you have a holistic thing like one of these documentation forms, like a guideline, like a protocol, there are a whole set of rules that are being assembled together and it's important for you to be thinking about how to make sure that all those rules are consistent with each other. Most EHR vendors do not give you tools to ensure that you are consistent across rules. So, you have to think about how to do that on your own. So, I do want you to be thinking about it either if you're going into this field for real or if you're a consumer you turned around and you ask the folks doing your knowledge management. How are you making sure that the rules are consistent with each other? So, I've shown you some graph you can see in the curves that formulas probably make a better a better basis of rules then all those rectangles that I'm showing you. We'll be talking about formulas in the next module when we talked about how to put knowledge into a system but we're going to spend the rest of this module on our issues associated with the language of rules. But what are the words, terms, and concepts we put into the rules. So, it's thinking coherently and inconsistently. So, that's going to be data dictionary, taxonomy, value sets, semantic networks, and ontology. Our next session we'll turn to the next two items.
