I’ll be writing a lot about search filters in the near future. Before I do, I think it would be useful to give an introduction.
First I want to stress that this series will not deal with Google, Twitter etc. search filters. Although I might write about such filters on another occasion, this series is about filters for biomedical bibliographic databases, such as MEDLINE (PubMed) or EMBASE.
Below is a short Dutch presentation on search filters I gave at a symposium on search filters in 2005. (some slides don’t show well in Slideshare)
Search filters are predefined and pretested search queries designed to retrieve selections of records in specified electronic information sources. Usually they are created by librarians, but they can be run by clinicians and researchers as well.
Why are search filters useful?
It is increasingly difficult, especially for the busy clinician, to find the information he/she wants in a database like PubMed. Filters can help to narrow down the search. In this way you can reduce the number needed to read (i.e. in a quick search) and/or increase the number of relevant papers (i.e. for a systematic review).
Different classes of filters:
- Subject vs Methodological Filters
- Subject filters are search queries for a topic, i.e. the BestBets paediatric filter (to search broadly for articles on children) or the build-in toxicity filters in ALTBIB. Both are search strategies for PubMed.
- Methodological Filters, Qualitative Filters or Hedges are search strategies designed to find studies with a certain design or methodology. PubMed has build-in methodological filters, the so called Clinical Queries for domains (therapy, diagnosis, etiology/harm, or prognosis) and for evidence based papers (like systematic reviews and guidelines, see Systematic Review subset in Pubmed).
- Sensitive vs Specific Filters
- Sensitive filters are broad filters, designed to find as much relevant papers as possible, often at the cost of much ‘noise’.
- Specific filters are designed to find a small set of very relevant papers, with the risk of omission of a considerable number of relevant papers.
- Short vs Long Filters
Filters may be simple -even consisting of one single term- or may be complex. They can comprise keywords, text words or both.
- Database and search-interface dependency
Filters are usually designed for a specific database and interface. Not every filter that has been developed can be directly translated into another database/platform because of different keywords terms, hierarchy of keywords, structure and commands. For instance in EMBASE Case Control Study (which is not strictly a controlled study) is a Narrower term of Controlled Study, together with Controlled Clinical Trial. In MEDLINE Case-Control Studies‘ is considered an epidemiologic study, whereas Controlled Clinical Trials is a Publication Type. Note that the MESH is in the plural form: ‘Case-Control Studies‘
The OVID platform allows separate searching of the abstract field (command: .ab.), whereas PubMed has no separate command for this ([tiab], title and abstract). Adjacency searching and non-explosion of subheadings (qualifiers of a MeSH term) is also possible in OVID, but not in PubMed.
- Time dependency
The performance of a filter may change over time because other terms may prevail or database keywords may have been added, removed or changed. In PubMed for instance the MeSH Randomized controlled trial has been changed in the MeSH Randomized controlled trial as a topic.
- Subjective vs Objective Filters.[1,2]
- The 1st generation of filters are subjectively derived, based on the expertise of the searcher.
- 2nd generation is also subjectively derived, but then tested and validated against a gold standard, i.e. a known set of relevant records, to determine the effectiveness of the filter at retrieving relevant records.
- 3rd generation involve objective approaches to filter design (e.g. frequency analysis or logistic regression). Search filter is tested on an independent set of known relevant records (gold standard).
Whether filters are broadly applicable to different clinical areas will depend on the choice of the golden standard (subject, publication year, size) and the presence and composition of an extern valididity standard (a set of records different from the records used to develop the filter, against which the developed filter is tested)
Performance measures (for 2nd and 3rd generation filters), Figure adapted from 
Sensitivity: The number of relevant records retrieved by the search filter as a proportion of the total number of records in the gold standard.(A/A+C)
Specificity: The number of records that are not relevant and are not retrieved as a proportion of the total number of records.(D/D+B)
Precision: (or positive predictive value), fraction of returned positives that are true positives. (A/(A+B)
Is there any difference between search filters and limits?
Not really, both are search terms that can be used to narrow the search. In PubMed however limits usually consist of ONE single MeSH-term (MEDLINE Subject Headings, i.e. key words assigned by MEDLINE-indexers), thus these limits will miss recent papers that have not been indexed by MEDLINE. Therefore it is often safer to use a broader filter or no limit at all.
For instance consider the search tinnitus AND behavioral treatment (set #1 in the Fig. below). This yields 175 hits.
Most people, if they want to find the best individual studies limit for RCT, i.e. the subject search is combined (ANDed) with “randomized controlled trial[Publication Type]“. This yields 23 hits (#2).
However if they would have combined ther search with the narrow therapy-filter (#3) they would have found 3 extra hits, two of which being RCT’s, but still in process and not indexed:
1: Weise C et al Biofeedback-based behavioral treatment for chronic tinnitus: results of a randomized controlled trial.J Consult Clin Psychol. 2008 Dec;76(6):1046-57.
2: Kaldo V, et al Internet versus group cognitive-behavioral treatment of distress associated with tinnitus: a randomized controlled trial.
Behav Ther. 2008 Dec;39(4):348-59. Epub 2008 Apr 20.
The narrow therapy filter has found these very recent articles, because it not only searches for randomized controlled trial[Publication Type], but also for randomized AND controlled AND trial in title and abstract.
The broad therapy filter (#4) searches for clinical trials in general and finds many clinical trials that are not RCT’s.
The narrow (#5) and broad (#6) Cochrane RCT-filters are highly sensitive search strategies meant to identify randomized controlled filters for a Cochrane Review (see 10).
Thus for a quick search for relevant papers search #3 (narrow therapy filter) is most optimal.
- Limpens J [ppt] Introductie Zoekfilters 2005 (Dutch)
- Booth A (Scharr) [ppt] Quality Search Filters at http://www.le.ac.uk/li/lgh/library/ABooth.ppt
- Jenkins M. Evaluation of methodological search filters–a review. Health Info Libr J. 2004 Sep;21(3):148-63.
- Glanville J, Bayliss S, Booth A et al So many filters, so little time: the development of a search filter appraisal checklist.J Med Libr Assoc. 2008 Oct;96(4):356-61.
- UBC Health LiBrary wiki: Systematic_review_searching and Filters (ie.hedges)
- Haynes RB, Wilczynski N, McKibbon, KA et al. Developing optimal search strategies for detecting clinically sound studies in Medline. J Am Med Inform Assoc. 1994 Nov-Dec;1(6):447-58.
- PubMed Clinical Queries
- Systematic review subset in PubMed
- McKibbon, KA, Wilczynski, NL, Haynes, RB Retrieving randomized controlled trials from medline: a comparison of 38 published search filters. Health Info Libr J. 2008.
- Post: New Cochrane Handbook: Altered Search Policies.