Welcome back. We are now on Section C of the third part of this presentation. And in this section, we're going to cover How to Build a High-Quality Search Strategy. I'm going to use the Medline PubMed database as an example. And show you how to find medical subject headings and other elements from a PubMed record. One of the things you're going to be asked to produce in the class is a table of the mesh and keywords that you use and we'll go into that. In building a high quality electronic search strategy, one of the guiding principles is being able to replicate it, as in any other experiment. So you want to record it in a way that nails it down. And I'm going to show you some of the tools that you can use to do that. I'll introduce you to the idea of study design filters. So if you're looking for randomized trials there are concepts that you can add to your search that will help focus in that area. And we will also introduce you to a nice simple tool for doing a double check on your search when you're done. So the process that's suggested for developing a MEDLINE Search Strategy for a systematic review for this class in particular is to start with a relatively simple search. You have a review that you're starting with that will have some citations in it. But from that, take terms, create a relatively simple strategy and retrieve some additional reports. Analyze them as in pull out the medical subject headings and the keywords from those studies and create a table. And identify the ones that look like they're going to be the most useful to your search. Then revise the strategy using those terms and rerun it. This looks like a nice linear set of steps but I think what you're going to find is it's pretty iterative. You do a search, you find things, you try it out. You look again. I'll say again, if you can remember to document what you're doing at this point, you will save yourself repeated effort. Even the terms that don't turn up to be useful, note that and then you want, a day or two later, go back and say, oh, did I use that or not? Once you have got the optimal search strategy, you run that, retrieve the citations and import them into end note and go on to the next database. So in developing your search strategy, what you need to do is take your research question and break it into concepts. The PICO format that many of you may be familiar with is helpful here. That stands for Population, Intervention, Comparison, and Outcome. So here's an example that has to do with age related macular degeneration, and whether Intravitreal injections of Lucentis are better than Avastin to prevent vision loss. So the slide gives you how you would take that question and break it into those four concepts. Once you have done that then you can start translating this into a format that a database can understand. So here you see the four concepts for population with synonyms OR-ed to each other. And then at the end you and that with the next concept, the series of synonyms that represent that concept. In doing a search for a systematic review, less is more. The fewest number of concepts that you can use, and have a manageable set of results, which is still in the thousands, the better. It's rare that for systematic review you include the outcome terms, because it's so easy simply to overlook one of the terms or words that would be used to identify it. And once you've got that together, then if there is a tested field study type filter available, you add that into the search. What I'm talking here is basically logic. And I don't know of any large databases doesn't have that available or doesn't use that as a basis for searching. So AND gives you an intersection between two sets. In this case, macular degeneration and intravitreal injections. Or expands what you have by using synonyms or other terms or ways that an author would describe what they're doing in an abstract or title. NOT is a tool that I use when I'm assessing a search. I might add in a term, and it says, oh, it's added 50 citations. Well, let me see what those are. I can take the two searches and NOT out everything except those 50 citations and see if it was really helpful or not for example. But I would not use that in the final search strategy. Let's move on to PubMed itself. And let me show you what this might look like in practice. So you come to PubMed just using the terms that we had in the previous slide. You type those into the search box up at the top and you get a series of citations. You see we got 400 hits here. So, what I typed in is at the top. But the search that actually got done, is what's down at the bottom right. The Search Details. And if we go underneath that box, there's a see more. You'll see the whole thing. So, even though all I typed in was macular degeneration. PubMed, behind the scenes, is trying to help me. It went ahead and ran those terms against its list of medical subject headings and said, oh there's a subject heading macular degeneration. I bet that's what you wanted. And put it in the search for me. Now in this case, it worked just fine. It is really a good idea to check the search details when you do a search, because it usually gets it right, but sometimes you may get some things that you really don't want in a search strategy. So if I were to click on one of the citations that was retrieved from that search, I might see something like this. Here's a record in abstract view. You can find keywords to add to your search from the abstract and the citation. You can look to the related citations on the right hand side of the screen, to do some Snowballing of articles. If you scroll down you're most likely to see this what you see here. A single line with a plus sign that says publication types MeSH terms and other terms. If you click on that plus sign, what you will get is a list of the publication types that have been applied to this article and the MeSH terms that the indexer applied. So this is a great place to identify the terms that were used for the specific articles. Back at the abstract view, on the left hand top side, you have the opportunity to change the display settings. If you go from Abstract to Medline, let me show you what you'll see. I've been talking about searching a record. This slide shows you the structure that the record is actually found in the database. When we search for MeSH headings, we search in the fields that are labeled MH. That lists on the left are what are called Field Codes. And they identify the specific fields where particular information is put. So, if somewhat about that you see TA, that's the journal title. There's a field for the TI for the title and so on. When you do keyword searching, you pick the other fields, other than the MeSH terms, that you want to search, to find these words. Another way to find MeSH is to go straight to the database of MeSH terms that the national library provides for you. So if you go to the home page of PubMed and click on the link on the right, that will take you into the MeSH database. Now if you put in search terms you're searching in the vocabulary, not in the articles. And what you get back are a list of any entries for terms that matches your search. So here we have, again, the entry for Macular Degeneration. You click on the title and it takes you into the record that you saw previously for Macular Degeneration. If you see at the bottom of the slide, I said that the search will search on everything unless you tell it otherwise. You see where it says do not include MeSH terms found below this term, you click on that box, and that's one way to limit the search. Here's the bottom of the screen again. And this is an example of the kind of table that you'll be asked to put together with your collected MeSH terms and keywords. And what's being suggested is that you use the rows for the individual papers. Use the columns for all of the various terms. And then identify which papers have what terms. And that should help you get a overall picture for which terms are going to be most helpful to you in building your search. I showed you the slide a couple of slides back that had the field tags. Here's a list of some of the most common ones. At the bottom is the URL where you can find the complete list of fields that you can search. You can also get to it from the Help button that's found on every page in PubMed. When you do keyword searching, in general, starting with the text word field tag is a good compromise. That includes most but not all of the fields that have text in them. For example, it excludes the affiliation field. So if you're searching for something in Baltimore, you wont get every institution that happens to be in Baltimore in your search. It also doesn't include individual author so you won't get Mr. Baltimore's papers. If that still pulls up too many false results, you can narrow it to just title and abstract. In some cases, using all fields is actually a good idea. Particularly for topics that are not sort of core medical topics or less well indexed. So once you have done the first iteration, done a simple search, collected your MeSH terms, now is the time to refine your search strategy. Add in the additional terms that you found. Pay attention to plurals. You can use, at the very bottom of the slide you see Truncation. That may be one way to take care of those variations. Think about Abbreviations that are specific enough to your topic to be useful without pulling up again, false hits. Remember spelling variations. British spelling versus American spelling. I have run into snags when not remembering that anemia has ae in various places. Other questions that come up in building a search include whether it's okay to limit say by language or by year. This slide shows some results of studies that have been done on systematic reviews. Egger for example looked at papers that related to the same research project that German researchers published in German and English simultaneously. And in this case, more positive findings tended to be published in English than in German. So if you were only to include English, you would be unintentionally biasing your results towards the positive. The same trend was found in complementary medicine areas and studies from China, Russia, and Taiwan. Juni did a study looking at Cochrane reviews in comprehensive searches and that was not found to be the case. Additionally, just in general, be very cautious about using limits. You need to be able to justify any limit you add to your search. Date limits are generally only used if, for example, a drug was introduced in a specific year or a particular disease emerged at a particular time. Now a few words about finding or use of search filters for Study Design. We've been talking a fair bit about developing concepts, that represent the actual content of your search. I know that some of you are going to be doing reviews of randomized trials, and others of observational studies. There are some filters developed that have been tested in MEDLINE and NBASE for identifying randomized control trials. What you see on this slide is the Cochrane Highly Sensitive Search Strategy. If you go to chapter six, you will see a couple of different iterations of that. And, you can read about which ones would be useful there. If there are tested filters for the databases, you can add them to your search. Here are some examples of articles published that have done in research on MEDLINE, MBase, and other subject specific databases and optimal search strategies for finding different types of studies. This slide gives you a picture of one finished search formatted for PubMed. The top set of terms, word together, represent the population that's being studied. The bottom set are all ORed together and actually represent both at the intervention and the comparison for the example that we've been using. And after, as I've said, you get those subject specific concepts represented in the terminology. Then you can add study filter terms to that search. Margarete Sampson and her colleagues who are information professionals doing research on the quality of searches. Reviewed the literature on tools that have been used to evaluate the quality of electronics search strategies. They identified 26 and teased out from that seven key criteria that can be used to assess your search. It's a nice checklist, it's available with this set of slides. The PRESS stands for Peer Reviewed Electronic Search Strategies. Sampson and McGowan have also done research on the kinds of errors that creep into even published search strategies. One of the key areas that people tend to overlook, is the need to adapt your strategy. I've been using PubMed and Medline as an example. Once you've developed your strategy, then when you move to the other databases, you're going to search. You need to take that and actually find the controlled vocabulary for that database, find the appropriate truncation symbols and other items to make the search work effectively in the other source. In summary, this section of the presentation has covered what it takes to build a high quality electronic search strategy that you can then record and have as part of your systematic review report. You start by developing your search. Take your question and break it into concepts using the PICO format. Identify synonyms and control vocabulary. Use the tools that are available from the database, both Boolean Logic and Controlled Vocabulary. Remember, this is going to be iterative. You're going to be revising your search strategy until you come to one that is optimal. And remember when possible to remember when there are tested study design filters available to use them. And we've introduced a tool that you can use to double check your work when you're done, the PRESS checklist. In the final section I'm going to cover a few more words about documenting you process and conclusions.
