Case study of a search
The aim of your search is to create a strategy to identify all the relevant and high-quality material to answer your question (often referred to as search sensitivity) without identifying any irrelevant, low-quality material (often referred to as search specificity).
The ideal search is both sufficiently sensitive and sufficiently specific.
The following is a sample Medline search (designed to run on the OVID Online search interface) to search for a question on empirical treatment in adults and children with suspected bacterial conjunctivitis:
Lines 1–33 list terms designed to retrieve the correct study type. In this example, systematic reviews (1–25) and RCTs (26–33)
1. review.pt.
2. (medline or medlars or embase or pubmed or cochrane).tw,sh.
3. (scisearch or psychinfo or psycinfo).tw,sh.
4. (psychlit or psyclit).tw,sh.
5. cinahl.tw,sh.
6. ((hand adj2 search$) or (manual$ adj2 search$)).tw,sh.
7. (electronic database$ or bibliographic database$ or computeri?ed database$ or online database$).tw,sh.
8. (pooling or pooled or mantel haenszel).tw,sh.
9. (peto or dersimonian or der simonian or fixed effect).tw,sh.
10. (retraction of publication or retracted publication).pt.
11. or/2-10
12. 1 and 11
13. meta-analysis.pt.
14. meta-analysis.sh.
15. (meta-analys$ or meta analys$ or metaanalys$).tw,sh.
16. (systematic$ adj5 review$).tw,sh.
17. (systematic$ adj5 overview$).tw,sh.
18. (quantitativ$ adj5 review$).tw,sh.
19. (quantitativ$ adj5 overview$).tw,sh.
20. (quantitativ$ adj5 synthesis$).tw,sh.
21. (methodologic$ adj5 review$).tw,sh.
22. (methodologic$ adj5 overview$).tw,sh.
23. (integrative research review$ or research integration).tw.
24. or/13-23
25. 12 or 24
26. "randomized controlled trial".pt.
27. (random$ or placebo$ or single blind$ or double blind$ or triple blind$).ti,ab.
28. (retraction of publication or retracted publication).pt.
29. or/26-28
30. (animals not humans).sh.
31. ((comment or editorial or meta-analysis or practice-guideline or review or letter) not "randomized controlled trial").pt.
32. (random sampl$ or random digit$ or random effect$ or random survey or random regression).ti,ab. not "randomized controlled trial".pt.
33. 29 not (30 or 31 or 32)
Lines 34–43 consist of search terms chosen to retrieve the correct condition, in this case conjunctivitis
34. exp Conjunctivitis/
35. Conjunctivitis, Allergic/
36. exp Conjunctivitis, Bacterial/
37. exp Conjunctivitis, Viral/
38. exp Keratoconjunctivitis/
39. (pink$ adj3 eye$).tw.
40. conjunctivitis.tw.
41. (Inflam$ adj3 conjunctiva$).tw.
42. Keratoconjunctivitis.tw.
43. or/34-42
Line 44 combines the strategies for publication type and condition, leading to retrieval of studies that are of specified quality (line 25 for systematic reviews and line 33 for RCTs) and are relevant to the specified population of interest
44. 25 and 43
In order to make your search more specific, you can also add search terms for interventions of interest. This will save you time in appraising the results of your search. Adding interventions, however, reduces the sensitivity of your search. Indexing of papers can be variable, not always containing all the relevant intervention terms, so you risk missing some relevant papers. You could, for example, add the terms:
45. exp Anti-Bacterial Agents
46. (ocular adj decongestant$).tw.
47. (eye$ AND decongestant$).tw.
48. Sodium Chloride/ or Saline Solution, Hypertonic/
49. compress$.tw.
50. or/45-49
Where there are established subject headings for interventions of interest, it can be very useful to add them to make your search more specific. If you are unable to find an appropriate subject heading for intervention, however, it may be better to simply search on population terms alone and use your critical appraisal to narrow your search to relevant interventions. See below for further details on subject headings and text words. See Appraise the evidence for further details on critical appraisal.
Line 51 combines the strategies for publication type, condition, and intervention, leading to retrieval of studies that are of specified quality (systematic reviews and RCTs), relevant to the specified population of interest, and relevant to the specified interventions of interest.
51. 44 and 50
Subject headings and text words
Thorough search strategies use both subject headings and text word searches. Using subject headings that are part of a controlled vocabulary or thesaurus (e.g., MeSH as used in Medline, Emtree as used in Embase, etc.) are useful because:
- They enable the user to retrieve relevant references where the authors have used different terms to describe the concept of interest
- The title or abstract may not explicitly mention the concept of interest
- The subject headings may give additional information to that provided in the title and abstract
- There may be no abstract.
However, subject headings are not always applied accurately or consistently by indexers. Searchers may not choose all appropriate subject-heading terms. It is therefore also useful to use text words in searches.
For both subject headings and text-word searches, use the main condition terms and synonyms where appropriate.
In the above example, the subject heading for Conjunctivitis (line 34) has been exploded (note the ‘exp’ instruction in front of the term). Exploding this term picks up the more precise or ‘narrower’ subject headings below it, which have, for the sake of clarity, been included within the example (lines 35 to 38 comprise the narrower headings below the main heading of Conjunctivitis. Where these headings have their own narrower terms beneath them, they too have been exploded, i.e. lines 36 to 38).
This approach to searching (combining subject-heading searches with text word searches) is standard search practice within an evidence-based context.
Different databases often use a different controlled vocabulary, so you need to check and use the appropriate subject headings for that specific resource. Also, different search interfaces (OVID; Dialog; PubMed; Cochrane Library; etc.) may use different methods to access their controlled vocabulary, or different syntax to denote when to explode or focus a subject heading, so it is important to check how to use and apply controlled terms appropriately.
For further explanation and advice on using controlled vocabulary when searching a database, consult your Library or Information Services staff, who should be able to assist.
Search commands
There are various search techniques to enhance text word searches. Truncation and wildcards are used to broaden the scope of our searches. For example, truncating the word random$ (line 27) will retrieve studies that include not only the word random but also randomized and randomization.
On occasion, if appropriate, we will also use techniques that narrow the focus of our search: for example, frequency and adjacency operators. This helps us to limit the number of irrelevant references we retrieve. (e.g. use of the adjacency operator ‘adj3’ in line 41, only searches for the term inflam$ within 3 words of conjunctiva$).
Different search interfaces (OVID; Dialog; PubMed; Cochrane Library; etc.) may use different syntax to denote when to apply a specific search command (e.g. truncation might be signaled by ‘$’ or ‘*’ or ‘?’ or something else) and some search systems may have commands available that others do not, so it is important to check what search commands are available and how to apply them, before you finalize your search strategy. For further details on the available search commands in a database and how best to use them, check the database help function or consult your Library or Information Services staff, who should be able to advise.
How BMJ Knowledge Center searches have been designed
The pressures and realities of updating products (and/or completing ad-hoc projects) means that our information specialists design any required search strategies to be pragmatic — processes and results need to be manageable within the time and resources available. Within this context, however, our search strategies must also be reasonably robust and comprehensive — when being used for evidence-based publications or reports.
When defining our inclusion criteria for any required BMJ Knowledge Center systematic searches, we use a PICOT structure; however, depending on the product and the number of references likely to be retrieved, the search strategies that inform a clinical issue may be kept quite broad, including only terms for the relevant condition/population.
Obtaining Further Support
Search strategies often need to be tested and refined in order to identify the optimum approach, experimenting with different terminology or search commands, in order to find relevant results in a sensible timeframe, for any particular project. If your institution does offer help and support with literature searching (usually via the Library or Information Services team) it is worthwhile utilizing their assistance and guidance, as this can be extremely helpful and ensure the search phase runs smoothly. Library and Information Services staff can also help in the acquisition of full-text articles and the use of reference management software (e.g. EndNote, Mendeley, etc.) so records of relevant papers can be easily managed and referred to again when needed.