MLA 2006:
Questions and Answers
[Editor's Note: Several of the Questions and Answers pertain to the subheading consolidation proposal. Please see decision].
- NLM Online Users' Meeting
- Meeting Remarks - David Gillikin, Head, MEDLARS Management Section
- PowerPoint® Presentations for Meeting Remarks, MedlinePlus®, DOCLINE® and MeSH®
Questions and Answers- NLM Theater PowerPoint® Presentations
- NLM Update PowerPoint® Presentation
Question:
I would like to get some input from the audience. As professional searchers, how many of you, when you are doing a comprehensive search, limit your search to just heading/subheading combinations? I would respectfully submit that the subheading combinations are for precision searching -- when you need to go through thirteen million PubMed citations and find a few things that have a specific role. If the general user does not use subheadings and the professional users do not use headings/subheadings for comprehensive searching then there is no reason to eliminate them even if some are not assigned by the indexers. Doing so will eliminate the precision for those of us who need to find a few ideal citations for our users in a specific area. Could you please explain why you are doing this when it doesn't seem to have anything but a negative effect on the searching capabilities of the database?
Answer - Dr. Nelson:
I think I have explained a number of good reasons to revise the subheadings. I didn't suggest eliminating the subheadings. I explained about skewed meanings and inappropriate use of subheadings. I don't think you are completely accurate in saying that the only reason not to do it is to improve precision. That is not the case; we have other problems and I addressed those issues. I talked about representational integrity, about special meanings. These are serious issues.
Question/Comment:
This seems to be a wonderful opportunity to go evidence-based. I am wondering what thought has been given to some form of test database, focus groups or data-mining -- simulations that could be used to test the questions that are coming up. We can have different layers of users making sure that all of these changes proposed have good effects or optimal effects. Your work is extremely important but to a huge audience. If you don't stratify your users all of these questions don't make any sense. I think that we need to get an expert group together representing different stake holders in order to settle some of these questions before the work gets done. I know what the costs are and I would really like to see that we do this in a cost-effective way.
Answer - Dr. Nelson:
There are a lot of serious questions about how to change this. There are fundamental problems in the whole information retrieval paradigm of understanding when you have found something that's relevant. There are a lot of problems with trying to test this out. The only person who can really understand whether something is relevant to a search is the user. You can know something well, and, therefore, the article that you are looking at is not helpful at all; it's essentially irrelevant at that point. There are other times when you don't know very much so that an article is quite helpful. Even measuring relevancy which is fundamental to information retrieval is a difficult task. One of the things that I have done is look for and ask what other information is available. There is somebody at NLM who is looking at developing semi-automated methods for assigning subheadings. One of the things that person is doing is looking at the indexing consistency data to find where subheadings are frequently confused, to find which ones have been used consistently with a much higher standard than the average. Right now all we have are average numbers, but within that, I am sure we are going to see a broad spectrum of subheadings that are easily confused as well as ones that are highly consistent. I think that is probably the best thing we can go on right now in terms of evidence.
Question/Comment:
As a great MEDLINE® fan for many years, my comment relates to the NLM Technical Bulletin of February 2006, where it was announced that the PubMed search tag [AB] was removed. [Editor's Note: Please see Technical Notes. NLM Tech Bull. 2006 Jan-Feb;(348):e1.] for the full announcement.] I fully support the rationale behind removing the ability to tag with AB because it wasn't limiting the search to the abstract field at all and it seemed very sensible to remove that -- it was never a supported field limit. I think it was what you described as an alias for text word searching. Nonetheless, what my request is is that you actually instigate the ability to search in the abstract field. You are probably aware that the functionality exists at the moment to limit to both the title and the abstract concurrently with the [TIAB] tag, but for those of us who are interested in designing highly sensitive and highly precise search strategies, it is necessary to limit sometimes to the abstract only. For those of you who have seen the current issue of Journal of the Medical Library Association (JMLA), we [Cochrane Collaboration] just have published a new highly sensitive search strategy for identifying Randomized Controlled Trials (RCTs) in MEDLINE. This relies on being able to search certain words in the abstract only, not the title and abstract. And, for example, that search strategy has improved the precision of the clinical queries that you currently have for the therapy filter from something like 25% to 40% precision, but you can only do this if you can search in the abstract only. We did the research in Ovid®, and for the benefit of people who don't have access to Ovid we translated this into PubMed. And of course we made the mistake of translating limits to abstract to the AB in square brackets. So I would like to request that you have a look at adding the ability to search only the abstract field.
Answer - Mr. Gillikin:
We examined a number of searches using a "sensitive" search strategy to see if an abstract-only search tag in PubMed would yield results significantly different than using [TIAB]. We did not see this. We have, therefore, decided against adding a new search tag at this time.
Question/Comment:
I would like to say that from all of the subheading announcements that we saw prior to this meeting, I never got the impression that we should go through each subheading and make a case for or against its relevance or why it made sense. From part of the discussion it sounds like maybe that's the kind of feedback that you would really prefer. Something that caught my attention from the letters that we saw posted to MEDLIB-L was that some people were okay with not having the secondary subheading but then I read that one cancer librarian was very concerned about not having it. How you balance that really depends on the kind of searching you are doing and also the utility of the different subheadings that are available. The second question that I have is what happens to the system? How will changes in 2008 impact older citations? Will just everything from 2008 forward have the new subheadings? When I conduct a search will I need to take into account if it's abnormalities or birth defects? Do I also need to search with abnormalities for citations prior to 2008? Doesn't this make it more confusing and harder for people to find things when you have to know the cut off in your head to know how to construct your search?
Answer - Dr. Nelson:
Answering the second question first: Just as you do not have to change the MeSH® heading that you use to correspond to which year MeSH headings were assigned to MEDLINE citations, the same process will be applied to the subheadings. We do citation maintenance. It's a process which is not really well described. Susanne Humphrey wrote a paper about it a good number of years ago [Editor's Note: Humphrey S. File maintenance of MeSH headings in MEDLINE. J Am Soc Inf Sci. 1984 Jan;35(1):34-44. PMID: 10299353]. If you go to the MeSH Web site there is some other information including an XML of the actual transactions that we follow every year to do citation maintenance in order to maintain the upcoming year of MeSH. I expect to do exactly the same thing with subheading revisions so that, no, you won't have use two different systems of subheadings.
In reference to the first comment, the article that is on the MeSH Web site addressed the idea that I wanted to hear comments about usability, reproducibility and understandability. I really do welcome those kinds of comments. I look at it from the point of view of a professional who works in vocabulary development and I say that this system needs repair and needs to evolve. But for resolution, I look at the comments. What I do most of the time is look at requests, comments from the Index Section, comments from the catalogers, comments from people who deal with the users, and comments from individuals who work with the system. I try to balance all those things out. I will read the comments carefully and consider them. I want you to know that I do have an open mind about where we are. I was deliberately provocative in the draft that was sent out. I wanted to make people stop and think about everything being on the table. Let's think about what it is that we really want, what is it that we really need, and how do we make what we have better?
Question:
And so does that mean that you're open to adding subheadings instead of just eliminating them?
Answer - Dr. Nelson:
If you can make the case that it is useful, understandable and can be done reproducibly, then I am open to it. However, the more subheadings you have the less likely it is to be reproducible.
Comment from David Gillikin:
For those of you who have been sending comments to MEDLIB-L, please do not assume that NLM is reviewing comments from MEDLIB-L. If you have a comment, please go to the MeSH homepage and follow the instructions for sending in comments. They then are forwarded to Dr. Nelson. [Editor's Note: The deadline for comments was June 16, 2006.]
Comment/Question:
I did send Dr. Nelson a comment on /chemically induced with respect to Lupus and I'm happy to see that's been moved over into the save column. Thank you very much. My question has to do with searching. Are there any plans to make true adjacency searching possible in PubMed and if not, why, when so many other databases allow you to do this?
Answer - Mr. Gillikin:
While there are no plans currently to add another method for adjacency searching, we are always interested in examining how to make PubMed a better search tool. It will help us considerably if you can send us searches that do not work for you using the current method of phrase searching. Please send your search examples for adjacency to: custserv@nlm.nih.gov.
Question/Comment:
My question is actually more of a request. We are finding that our clinicians and medical students are using Google™ Scholar. It's easy for them to get to Google Scholar when they are on-site -- our open URL links show up. We have open URL in PubMed as well and we use LinkOut®. I find that Clinical Queries is very popular among these clinicians and I would like to be able to have a direct link either from the Library Web page or hopefully some of the hospital Internet sites directly into Clinical Queries that would access our open URL links and our LinkOut. It's not realistic for everybody to go through and set up My NCBI filters throughout the hospital on the hospital computers so that those links are available. I'd like to see some sort of capacity for an individual link like we have through the Library Web site for our holdings to be available through Clinical Queries.
Answer - Mr. Gillikin:
Libraries can display the LinkOut icon through the Clinical Queries page. To do this, enter Clinical Queries using the following URL:
- //www.ncbi.nlm.nih.gov/entrez/clinicalx.cgi?holding=NameAbbr
- Replace NameAbbr with the provider's LinkOut User Name
Details can be found at the LinkOut homepage. Click on Libraries and then on Localize PubMed.
Additional information can also be found in the Technical Note, Displaying LinkOut and Outside Tool Icons When Using Single Citation Matcher and Clinical Queries. NLM Tech Bull. 2005 Jan-Feb;(342):e1.
Question/Comment:
I do a lot of PubMed training and I know that the National Training Center & Clearinghouse (NTCC) staff in its 8-hour PubMed class provides a brief overview of MeSH. As I travel around and train I find that most people are self taught when it comes to the MeSH vocabulary and I think that may be one of the problems. I think it would behoove the National Library of Medicine to develop a training class specifically on MeSH because at this time people are getting just a few minutes in the NTCC class. When I was reviewing comments about the subheadings I noticed that people were unaware of the fact that the subheadings are arranged in families and that they explode automatically in PubMed. This is basic information about using the MeSH vocabulary and it is not being effectively transmitted to people so a MeSH class would be helpful. This would be the logical vehicle to get the word out when the annual MeSH changes are made and instruct people on how those changes may effect their searching.
Answer - Mr. Gillikin :
Thank you for the suggestion. The current PubMed training class devotes one and one half hours to MeSH. In addition it is frequently discussed throughout other portions of the class. Also, please see the 12-minute video Branching Out: The MeSH® Vocabulary and the newest MeSH instructional resource, the Basics of Medical Subject Headings (MeSH).
Question/Comment:
I would like to see healthcare administration journals added as a separate subset in PubMed. We have a graduate healthcare administration program and we don't have many of the clinical journals in our library. I would like our students to be able to identify the journals that we are more likely to have and to be pertinent to their search.
Answer - Mr. Gillikin :
NLM doesn't have any plans to develop a healthcare administration subset for PubMed; however, if your library is a LinkOut participant you can establish a filter for your library's holdings. This filter can be part of a shared My NCBI account. The following links will take you to more details about LinkOut and creating a filter for a shared account:
- The LinkOut for Libraries Web page:
//www.ncbi.nlm.nih.gov/entrez/linkout/doc/liblinkout.html - Library LinkOut Filters Quick Tour:
//www.nlm.nih.gov/bsd/viewlet/myncbi/library_linkout_filter.html - NLM Technical Bulletin article:
Share My NCBI Account Settings. NLM Tech Bull. 2005 Jul-Aug;(345):e7.
Comment:
I would like to thank you very much. I'm a huge fan of PubMed and enjoy the resources and I appreciate all your hard work.
Question/Comment:
When I teach doctors or students I tell them to be aware when using MeSH because all in-process and publisher-supplied citations are not indexed. I tell them to use a keyword search term instead of going to the MeSH database. I say that because we tend to teach doctors to use MeSH and we emphasize it. If you want the most recent citations to the hot topics or those articles that just came in (and NLM does a wonderful job of putting them in for us to access -- there are a large number that are supplied by publisher and are available only electronically) these are not indexed yet as well. When I look in the Details I also realize that when you use MeSH you don't cover some of the terms that are being used by the author that are not really matching with the MeSH. I find it makes a big difference. Librarians might want to be aware of that.
Answer - Mr. Gillikin:
That is a good point. In fact, most PubMed searchers use "Google-style" subject searching, i.e., they enter the topics of interest and do not specify terms as being MeSH. Example:
- behcet disease small bowel
Users are comfortable searching using their own terminology and while many seem to appreciate the value of MeSH vocabulary and NLM's indexing, they do not rely on it for searching. This is OK because PubMed has extensive mapping capabilities that allow it to recognize the user's terminology as entry terms for MeSH headings or subheadings and incorporate them into the search.
Answer - Dr. Nelson:
I think what you are talking about is trying to go from a MeSH heading to a query expansion to get other types of commonly used text words and phases. Is that essentially what you are suggesting?
Attendee:
Yes.
Answer - Dr. Nelson:
What you do need to know is that there is a mapping file that sits behind the scenes and maps free text searching to MeSH and it's quite a large file. It's about 500,000 lines, each line being a single mapping. If there are terms in the vernacular or colloquial terms or frequently used terms that you don't find getting mapped to MeSH headings, please send those to me as a suggestion. We have the ability to add things to that file. Going from telling someone to use a MeSH heading and then to pull in other text words I think will likely cause problems. Most experiments that were done in that regard have shown that it actually detracts from the quality of the retrieval.
Question/Comment:
I teach classes for nurses and anytime you type in hospice nursing you get breast feeding articles because of that mapping of nursing and breast feeding. I have sent several e-mails to the MeSH section and NLM with no response. I just wanted to let you know formally that I hope this can be changed.
Answer - Dr. Nelson:
I suspect that what is happening is that nursing is being mapped to breast feeding. In fact, that is a common use and there are many times in the mapping file where there are mappings to several different meanings. For example, "cold" gets mapped three different ways. The problem is your users want a specific meaning and they want to exclude the other meaning of the same word. Unfortunately, we can't do that because the next person that comes along is going to want the other meaning for that specific word and that's where it becomes a problem.
Comment:
Thank you very much for making available products like PubMed Central® and Bookshelf. They are wonderful.
Question/Comment:
I am a grassroots librarian now after having run some big libraries. I love that you added the capability to limit searches to PubMed citations having links to either full text or free full text. It is really wonderful for our small college because we have ten doctoral students who don't have the money to go out and get the full text and we don't have the money to pay commercial services so this is just a great capability that you recently added. [Editor's Note: For further information on limiting your search results to include full text see: PubMed® Limits Page Updated. NLM Tech Bull. 2006 Mar-Apr;(349):e2. ] [Editor's Note: At this point, the taping service failed to capture the remaining portion of the meeting. The following questions were recreated from staff notes.] I would like more Oriental Medicine titles to be added to the database. I would like training telecasts and more distance education.
Question:
Is it a rumor that MeSH was developed by linguists?Answer - Ms. Humphreys:
Yes, it is a rumor. MeSH reflects influences from a variety of disciplines (including linguistics), but the person most responsible for its development was former NLM Director Frank Bradway Rogers, M.D., M.L.S. Winifred Sewell, Ph.D. was its first full-time editor. Those who are interested in the early development of MeSH and the work that led up to the development of MEDLARS should consult Scott Adams' 1972 Janet Doe lecture, "The Way of the Innovator: Notes Toward a Pre-History of MEDLARS" published in the Bulletin of the Medical Library Association and available in PubMed Central at: //www.pubmedcentral.gov/pagerender.fcgi?artid=197742&pageindex=1#page
MLA 2006: NLM® Online Users' Meeting 2006: Questions and Answers. NLM® Tech Bull. 2006 Jul-Aug;(351):e3c.
