Yale MeSH Analyzer

So here is a familiar scenario: I’ve got a search topic, and a starting set of known articles on the topic. I’m pretty sure these exist and they are relevant. My goal, however, is to find additional or all relevant articles on my topic. So I decide to do a search in PubMed, and I start to poke around to find search terms. I PICO the question. I create a concept table. Fill it with terms I can think of. Put together a search strategy, and do my search. But wait, there’s one article in my starting set of known relevant articles that just won’t show up in my search result set. Why is that? What’s wrong with my search? Which important term did I not include?

Typically what I do next is to compare the missing article with the ones that were successfully returned by the search, and see what makes it different. I can jump back and forth among PubMed pages for those articles, and hope I don’t get confused about them. Or I can open multiple windows on my screen, and hope I can see each one of them clearly. Or I can take out a piece of paper, and write down keywords that describe each of the articles, and hope that they will make sense to me when I put them together later on. A more systematic way to do the comparison is to create a grid like this one, where each column represents an article. Here is the article that was not in the result set, and the rest of the articles were in the result set. For each of the articles, PMIDs and author-years are used as identifiers. I can compare the Medical Subject Headings, and look for clues of why the missing article did not show up in my search result set.

As you can see, to make the grid easier to scan, I’ve grouped and sorted the MeSH headings alphabetically, and it is clear that the missing article is indexed differently from the other ones. We call this a “MeSH analysis grid”, because a key feature here is that it allows you to compare Medical Subject Headings for the articles. This happens all the time. Now just think about the amount tedious work it requires to create a grid like this every single time. I’ll need to locate the PubMed records for each one of the articles, copy and paste the fields I want, and group and sort the MeSH headings alphabetically. Shouldn’t there be a tool that does this automatically? And that’s exactly what the Yale MeSH Analyzer does. It takes a series of PMIDs, and creates a grid like this one automatically. Let me give you a demo here.

You can access the Yale MeSH Analyzer at this URL, and here’s the home page. So I can just put in a series of PMIDs into this big box here. It’ll take up to 20 PMIDs. You wouldn’t want that many columns in the grid anyway, as it’ll require horizontal scrolling, and quickly become hard to scan. You can delimit the PMIDs however you want: commas, spaces, each one on its own line. It doesn’t care. You can literally copy and paste a paragraph of free text with those PMIDs in it, such as from an email message, and the MeSH Analyzer will scan for everything that looks like PMIDs. So these are the PMIDs I want to create a grid for. Below the text box, I have some options. For now I’m just going to take all the defaults and see what happens. Click “GO!”, and here’s my grid, which looks like the one I just showed you. That’s how easy it is. Behind the scene, the program identified the PMIDs from the text input, queried PubMed those records, extracted the relevant metadata fields, automatically grouped and sorted the MeSH headings alphabetically, and presented the results in a grid format to you.

You might ask why we didn’t choose to put every single MeSH heading in its own row. That would create way too many rows for anybody to easily scan. That’ll actually defeat the purpose of the grid. So we settled on this format: not too many rows, still easy to scan. Of course, not everybody wants the grid to be all the same, so we provided some options. These options allow you to tweak the grid the way you want. You can choose to display the subheadings in full, in 2-letter codes, or not display them at all. You can display the article titles in full, truncate them to the first 100 characters, or not display them at all. You can choose to show, or not show, the abstracts, author-assigned keywords, or the field name column. You can choose the output format to be a Microsoft Excel spreadsheet, or a simple web page, like the one you just saw. Finally, you can also ask your browser to remember these settings, so that the choices you make here become the default for you whenever you use this browser.

So let’s test this out. Let’s say I want my subheadings in 2-letter codes, article titles in full, abstracts, keywords, and I want my browser to remember my settings. Click “GO!”. And here’s the grid I asked for. So in a grid like this, I can not only find indexing discrepancies, but I can also read the full titles, the abstracts, as well as author-assigned keywords, to look for terms and phrases that might help me expand my search to get absolutely the best recall on that topic. Going back. If you just want the simplest grid with MeSH Headings only, you can suppress subheadings, titles, abstracts, and field names. This time I want it in Excel format. And I don’t want my browser to remember these particular settings. Click “GO!”. And sure enough, I got an Excel spreadsheet with exactly what I wanted. Up till this point, using this tool requires you to put in PMIDs into this form, but it doesn’t have to be like that. You can actually use this tool right inside PubMed. All you have to do is to go to our help page, scroll down to the bottom, and follow the directions here. Drag this link here to your browser’s bookmarks or favorites bar to create a button. So you can see I now have a button called “Analyze MeSH!”, right here. You only have to do this once. Now I can go to PubMed, and do my search. Of course, this is a “dummy” search. If you’re going to do a serious literature search, you strategy will be much more complicated than this. But, to demo this tool, let’s say these are my search results, and I want to create a MeSH analysis grid for the first a few articles. Select them. Click on the button I just created. And here is my grid.

One thing to note is that this button does not give you access to the options on the home page obviously. It’ll use whatever the browser has as the defaults, so if you have used the “remember settings” function before on this site, it’ll just use whatever is remembered last time. So, to recap, what can you use this tool for? Well, first, it can help you identify the reason why some articles are missing in your search result set, especially if this is a key article that you know exists. That could be due to possible indexing messiness. There can be inconsistencies, missing indexing terms, etc. That’ll become apparent very quickly if you have a grid like this. That could also be due to problems with your search strategy. You may not have picked all the right terms in your search. Possible but less likely, you may have an article that you think is relevant but really does not belong. It happens. In that case, this is a powerful, visual way to show you why. This can also serve as a “scoping search” tool, to help you find potential terms you haven’t thought of yet. The grid can include fields such as the title, abstract, author-assigned keywords, and you can scan those fields, and you may discover terms and phrases that were not in your search strategy. This will help you cast your net as broadly as possible to capture every single possible relevant article on the topic.