So I'll first sojourn into the land of languages and rules, is this data dictionary and taxonomy. Set of concepts and we'll start off with a data dictionary. So, we'll start off with something simple, a hospital formulary. A formulary is a list of those medications that if a doctor prescribes them, then they can be ordered. So, you can already hear the rule that right, if you order them then it can be prescribed or they can be dispensed. A nice simple rule. How does it work, well if the name of the medication is on the list, it goes ahead. So, the first question is, what name are we talking about? You can see in these two columns, we have two possible names as the generic drug name, and it's a trade drug name. The generic name is generally the name that the drug uses, whatever country it is, whatever phase of development it is. The trade name does depend on what company you're at. Maybe what country you're in. So, it's a little bit less stable than the generic name. The actual drug, well this, the drug is not the name, the name is a property of the drug or it's an attribute of the drug. With this distinction between a thing and it's attribute things will be coming back and forth too. But at any rate, we got a generic drug name, we have a trade drug name. The formula is actually a bit more complicated than that. So, let's take a look at this here. So, they are the full column names of this part of the formulary. Often they also can show what roles those different columns are playing. So, we already saw in left hand side, the names of drugs or strings or venture character CELEXA. They're playing the role of names, that we already talked about. Let me go ahead to few columns over towards his concept, where it says anti-depressant or cephalosporins, that's telling me what class of drug this is in. So, Celexa is an antidepressant. The second column, the formulary column which says whether it's in the formulary or it's restricted formulary, then is an attribute of the drug in this hospital. So one of the distinguished, both them are kind of attributes of the drug. But I think you get a sense that the concept column, is something that is more stable across, let's say, countries or companies. No matter what you call Celexa, it's still an antidepressant. On the other hand, the formulary column well, one hospital could call it, it's on formulary, the hospital may not have it on formulary at all. Now the hospital may have it restricted. So, you can see those, we call them business knowledge, because it's not an attribute of the drug per se, it's a choice that the hospital makes. Similarly the last column, where the information is coming from again, It could come from different sources, and then the middle column, is also called a business rule, but it's a different type of business rule, because it's written in English. In order to implement those rules, I would need a programmer to translate this English, into something computable. So we talked before about implicit and explicit knowledge. So, the name is kind of explicit knowledge, whether it's formulary, is explicit knowledge. It's local but still knowledge. The concept is knowledge and it is explicit. But this is a column of implicit knowledge, because the computer does not have direct access to the knowledge about the drug. So, strengths, weaknesses or data dictionary, strings it's simple, it's a table. I know how to deal with tables, computers know how to deal with tables, they have database technologies. So, that's perfect. Weaknesses. Well, the left-hand side, the if part is pretty lame right? It's if the drug has this name then, not much more sophistication than that. So, the entries might not be computable as we saw with that middle column. Therefore, the data dictionary may still have inquisite knowledge and exquisite knowledge. Might not take advantage of external standards, we saw that in the case of the concept column, we were using a consequent vocabulary from AHFS. But maybe not, that language of formulary, restricted formulary, I think they made up. That's something that's used all over the world. So, that's not a standard, and then the opposite is that they may be knowledge that, anybody looking at these words knows. Everybody knows that Cephalexin is a cephalosporin, but if you're in a data dictionary that does not have a column that Cephalexin is cephalosporin, the machine does not know that, and then if I say as a physician my patient is allergic to cephalosporins, this data dictionary does not help that patient, because there's no drug called cephalosporin. Therefore either I as doctor has to list every cephalosporin that is known or I need some other rule somewhere, or it's complicated. So, that's about the data dictionary. So, next we turn to taxonomy, and taxonomy is built on that notion we saw from AHFS. The notion that Celexa is a antidepressant. So, there's the notion of a hierarchy of concept. Celexa as a drug is a concept. I can talk about Celexa without looking at any particular instance of it, and I'm not looking at a pill bottle of it, I'm not looking at a patient who's on it, I'm thinking about the concept of the Celexa. So, in abstract, Celexa is a concept. The notion of antidepressants, is also a concept, and I can relating one concept that selects it to another, the antidepressant, I'm saying that the first one is a case of the second, such that generates a hierarchy. There's places to look up vocabularies and concepts, and that take advantage of hierarchies as we'll see in a minute. So, the NCI is the National Cancer Institute for the National Institutes of Health in the US government and they put together this term browser. So, it's a nice, it's a free tool, go looking up, at how, what concept do the different vocabularies and terminologies use, for a theme that I care about. So, here's an example, I care about a heart attacks, I wanted to know how does ICD-10 and how do SNOMED talk about the concept of heart attack. So you can see I checked those two check boxes, I said I click on search and I get a list of hits. You can see the first hit says that myocardial infarction, is the way that one vocabulary talks about heart attack. There are also a number of concepts I might not have thought about. Fear, heart attack, fear of having a heart attack, anxiety about having a heart attack, have to be honest, are those for the same vocabulary? Are those from different vocabularies? If they're from different vocabularies, notice that there are three different ways of in English saying the exact same thing. Then we get to these two scores. Where did that come from? I don't care about a score. Well, clearly, those scores involved the notion of heart attack, but clearly myocardial infarction is one event. So, we use the word myocardial infarction and go back to NCA terms and see what I get from there. So, here again, looking at just ICD-10 SNOMED, now I get a much longer list because all the machine is doing is saying, "Let me look at the character string M-Y-O-R-C-A-R-D-I-A-L, looking to see if there's any entries that have those characters in them and giving me this hint." Let's just focus on ICD-10. You see this list of concepts and there you see one of them non-ST elevation NSTEMI myocardial infarction. Let's look at that for a moment. You see this is what an entry looks like in NCA terms. You could see at the top is the whole name. You can see the code in ICD-10 is I21.4. You can see tabs have different attributes of the concept. Remember name wasn't attribute. There are other attributes. Here, we see that there are synonyms, and you can see acute subendocardial myocardial infarction etc. You can see that the semantic type is that's a disease. It is a disease or a syndrome. A kind of hierarchical notion right there. I'm clicking on relationships because I'm making the point here that a taxonomy is that relationships between concepts, and sure enough, you can it says parent-child relationships. So, the parent concept is ischemic heart disease. So, it's telling me that the NSTEMI myocardial infarction is a ischemic heart disease. You can see ischemic heart disease is a range of codes, I20-125. ISCHEMIC heart disease, as a parent has multiple children, not not just the NSTEMI but these other NSTEMI heart attacks. All those five concepts of sibling concepts. We use familial relationships a lot talking about these concepts. This whole notion of hierarchy goes back several hundred years. Carl Linnaeus in 1700s developed this notion of a hierarchy to locate a species in a larger context so that a species is a member of the genus class which in turn is a member of families, which is a member of order etc up the line. We use this categorization till today. If you see that we give a name of a specie, we'll say, Canis lupus. Canis is its parent, the genus of canis, and lupus is a species of wolf, and genus itself is a Canidae canis and tells me, "Okay, it's in family canidae, and it's probably the whole genus of canis, etc." So, this notion of a hierarchy is very natural. This defines taxonomies. We're talking about PubMed. I can put it a term and I can look for the MeSH heading, the medical subject heading. You can see that the term lives in a swell hierarchy, and actually civil hierarchies within the MeSH heading. What this means is that, if I put the word, let's say, anemia into my search, not only does it give me articles tagged with anemia, but it gives me articles tagged with all the children of anemia. So, it gives me a broad search, but it's also a semantically meaningful broad search. It goes beyond using the words themselves. If I'm in a free text Internet search engine and I take anemia, it might find sites with anemia misspelled, but it might not be sites that about red cell aplasia, whereas MeSH can direct me to such articles. So, how is all this used? We've talked about infobuttons before. These are links to information that should be useful to the user. So, the information that comes to the user is generally text. So that's implicit knowledge, but actually the infobuttons use explicit knowledge to make their way to the notes that it's given you. Let's see how this works. So, you can see the problem is for this particular patient includes the diagnosis of Wegeners granulomatosis, which is displayed to the user but I can click on that little button, that 'i' button, and what happens is a search is performed in Wegeners granulomatosis to PubMed. Then a set of articles is sent back to me as the user to look at and proves. How does this work? While the patient information is that string Wegeners granulomatosis, somewhere either within the open infobutton software or over at the NLM, that string is mapped to the MeSH hierarchy which then locates the MeSH term for Wegeners granulomatosis. It did performs this search using MeSH. You get a query result, and then it sends the result back to the infobutton manager, which then displays it on the screen. So, the two main steps is mapping to the taxonomy or anthology, we'll get to that later, and knowledge of workflow and interfaces, how should the information be displayed on the screen. Now, the mapping is using explicit knowledge. It's using the taxonomies, and it's using either going up and down that hierarchy. The knowledge of workflow and interfaces generally is implicit. It's in the programming of the interface. So, the strengths of a taxonomy, it's pretty simple to understand. We have notions of hierarchy, we know is-a pretty well. So, it's simple to use. The weakness is, it only can deal with is-a and generally, there's only one pathway to the top, whereas, the reality is more complicated.
