1. Defining the Search Question

Before starting, it is important to establish a search question. The PICO check list can help formulate a question:

  • Patient population or problem
  • Intervention or Exposure
  • Comparison
  • Outcome
By using this checklist, an appropriate query can be drawn up, such as those supplied by the WHO. For example, 'Should zinc be used in the treatment of acute gastroenteritis in children in developing countries?'. The population is children in developing countries with acute gastroenteritis, the intervention is zinc, the comparison is with those not treated with zinc and the outcome would be clinical improvement. It is instructive to underline the essential words and phrases of the query. In this example, zinc, treatment, acute gastroenteritis, children and developing countries.

You will find that from the outset, it is important to properly understand your question.

To this end it is a good idea to learn from reading a few recent review articles or textbooks about your subject to be able to understand the issues and controversies involved, especially if it concerns an area that is new to you. Try to find out the wider implications of the subject, eg. why it is important, epidemiology, public health burden, diagnosis, treatment etc. This will help you to understand when you read the relevant studies, you may discover potential synonyms for the literature search and it will hopefully give you motivation for carrying out the project

The starting point for literature searching is to use Pubmed's clinical queries, which is freely available from going to www.pubmed.com and clicking on "clinical queries".

Keywords, MeSH terms, synonyms and spelling

Keywords and synonyms

Keywords are often those established by underlining key components of your search query. They are essentially free-text searches that search that exact term and thus can retrieve articles in which the term is incidental. PubMed attempts to assign a MeSH (Medical Subject Heading) term to your keyword, though will not search for other more-specific terms that may be more appropriate. Consequently many useful papers may be missed if MeSH terms are not considered as well as keywords.

It is useful to keyword search when the term you are searching for does not appear in the MeSH database, or the term is new or highly specific. New literature may not yet have been assigned a MeSH term and hence would only be found by keyword search. With keywords, there is a requirement to think of all synonyms or use a broad enough term to cover the topic. Brainstorm (or use a thesaurus) for alternative terms. Synonyms can be included in the search query using the Boolean operator ‘OR’ (explained later).

Truncations

If the keyword should have variant endings – singular, plural, adjective – truncation can side-step the need to combine them all with ‘OR’.

With truncation, use the common ‘stem’ of the word – for example, child - followed by the truncation symbol ($ * ? # - varies depending on the database being searched). Thus child* could retrieve child, children, childs, childrens etc. There is a risk that truncation may yield false ‘hits’.

MeSH terms

Every article entered into PubMed / Medline is assigned a MeSH term by a third party – essentially allocating the article a topic heading. MeSH contains almost 17,000 terms. Each of these terms represents a single concept appearing in the medical literature. As important new concepts appear, a new MeSH keyword is created. When a new reference is added to MEDLINE, indexers review the article and then choose/add the appropriate MeSH keywords (usually 10 to 20) to represent the contents of the article. Using MeSH terms in the search means that articles can be retrieved by topic and bypasses the problem of whether the right term has been chosen with keywords. A useful technique is to search the database of MeSH terms available here .

Establish the correct MeSH term by looking at the definition, and then look at the MeSH tree of Subheadings to see where this term fits in. By looking up and down the Subheadings tree, it is possible to make your search term more or less specific. Using the example of gastroenteritis, gastrointestinal diseases offer a far broader search, whereas dysentery or gastritis would restrict the yield of papers.

Using the MeSH terms means the most recognised form of the term is used and enables greater control over your search. However, as mentioned above, newer articles or new terms may not yet be allocated a MeSH term.

For completeness, both MeSH terms and keywords should be used in the search strategy. The importance of this is illustrated in the example below.

Remember:
- Keywords and MeSH terms
- Related terms
- Alternative spellings (particularly between UK and US English)
- Synonyms
- Truncations

Some advocate the use of a mind-map / spider diagram or columns to expand on these major topics. This is a matter for personal preference.

MeSH vs. keywords: an example

There will be vastly different results achieved depending on whether or not keywords or MeSH terms are used. Below is an example where MeSH terms do not entrain crucial articles to answer a question.

Take the question “What is the evidence behind dexamethasone therapy in bacterial meningitis in developing countries?”

Most authors will use keywords but these often do not match up with the MeSH terms. Using only one set will not produce the same results. The red example below uses the MeSH terms in the clinical filter and yet even with broader terms does not entrain a number of important articles. The keyword search is in blue and one can see that once put through the filter it entrains 6 further articles, (including the seminal work by Dr. Liz Molyneux.)

This of course occurs if the author when using the keyword search thinks of all synonyms or uses a broad enough term to cover the topic. They also have to take into account all truncations (hence “countr*”). The results in bold are based upon PubMed clinical queries searching although other databases/information repositories need to be searched as well.

#1 Search "Meningitis, Bacterial"[MeSH] AND "Developing Countries"[MeSH] 105 articles

» This is then put through the Clinical Queries filter/other database search engine:

#2 Search (#1) AND (randomized controlled trial[Publication Type] OR (randomized[Title/Abstract] AND controlled[Title/Abstract] AND trial[Title/Abstract])) 4 articles

#3 Search bacterial meningitis AND developing countr* 216 articles

» This is then put through the Clinical Queries filter:

#4 Search (#3) AND (randomized controlled trial[Publication Type] OR (randomized[Title/Abstract] AND controlled[Title/Abstract] AND trial[Title/Abstract])) 10 articles

So why not use keywords alone? The answer lies that in the example above it is relatively easy to work out the terms that will get the correct relevant articles. This is obviously not always the case. Running a search for skin grafting and burns the MeSH terms turned up a number of extra articles compared with keywords….61 Vs. 24. This is because the MeSH term is “skin transplantation” whilst with keywords the reviewer has to include a number of terms other than skin graft, such as ;

    · Dermatoplasty
    · Dermatoplasties
    · Grafting, Skin
    · Graftings, Skin
    · Skin Grafting
    · Skin Graftings
    · Transplantation, Skin
    · Skin Transplantations
    · Transplantations, Skin

”Transplantations. skin [MeSH]” picks them all up.

Another reason not to use keywords alone is that for broader topics then a number of irrelevant articles are captured which defeats the purpose of the clinical filter somewhat.

Combining the search terms

The keywords identified in the last section should be combined to search the databases. This refines the search. Use of the Boolean operators 'AND', 'OR, and 'NOT' achieves this.

AND retrieves records containing both / all of the combined terms. In the figure below, the shaded area represents the papers found when A (e.g. gastroenteritis) AND B (e.g. zinc) are combined.

AND

OR retrieves records containing either term. This is useful when synonymous terms exist. In this example, A represents gastroenteritis and B acute diarrhoea. For this search the terms are interchangeable so can be combined with OR.

OR

NOT is a term of exclusion, retrieving records that use one term but not another. In the figure below, B could represent gastritis but NOT A, appendicitis. This risks excluding records that could be useful, shown in the overlap.

NOT

In this example then, a final search strategy could be:
Zinc AND (acute diarrhoea OR gastroenteritis OR infectious diarrhoea OR acute gastroenteritis OR diarrhea).
Further examples can be found in the ‘examples of search queries’ in the reviewer’s toolkit.

As the search strategy is for the WHO review, it should focus on children in developing countries. At this stage however, it is preferable not to restrict the findings.


Inclusion and Exclusion Criteria - Setting Parameters

Before searching, set out on paper what you would and would not like to include. Obviously, the key terms should be included. Decide what sort of papers you need - systematic reviews and randomised controlled trials are the order of the day for the WHO review, but would you consider cohort studies or retrospective reviews? Humans only? Paper age? Paper language? Certain journals? Treatment versus prophylaxis? How will you select you papers?

Setting this out at the start avoids ambiguity later.

 The most readily accessible biomedical database is PubMed. By using the PubMed 'Clinical Queries' or 'Special Queries' options, you can set up specific searches to identify systematic reviews or different types of studies. This includes searches of the Cochrane Database of Systematic Reviews. Systematic reviews of policy interventions (such as pricing of pharmaceuticals) can be difficult to find and other search strategies will be needed. The Campbell Collaboration has a database of existing reviews of effectiveness of social and educational policies and practices. When in doubt - get help.

It is possible to specify these parameters on some databases (e.g. OVID) with check boxes for English language and children for example.

It is important to try to include studies from the developing world as well as the more standard literature. Developing country journals are not well represented in PubMed and commercial databases such as EMBASE, CAB Abstracts. Regional databases (AIM, IMEMR, HELLIS; LILACS, WPRIM) grouped under the general heading of the Global Health Index http://www.who.int/ghl/medicus/en/ contain unique citations and in many cases full text articles.
Regional offices of WHO have supported the development of the these indexes to highlight the health research literature of developed world. The majority of journals indexed by Regional databases are not indexed PubMed and other databases. An information specialist (eg at the WHO Library) can also suggest other databases to search depending of the topic areas (Legal, Social Sciences, Political etc).

 Retrieval of 'grey literature' such as Ministry of Health reports, case studies and unpublished studies, is best done based on the results from journal article searching from PubMed, Regional databases and other sources. Terms such as the key authors and institutions can be used with Google or other search engines to identify grey literature cited on the internet. It is also important to scan key web sites individually as general search engines are not capable of retrieving all the relevant information on a web site. Obtaining training in using search engines is important to limit research results to pertinent information. The combination of efficient use of search engines and targeted website/authors is much more effective for identifying unique information than large unfocused searches. Personal contact with key informants will help to identify sources of information not found in the published journals or cited on website.

Limiting or Expanding Search Results

Your strategy will almost invariably retrieve too many, too few or irrelevant hits at first try.

If too few hits:
  • Use truncation
  • Use a thesaurus - ensure all relevant synonyms are included
  • Spelling - check, or try an alternative
  • Check terminology
  • Combine keywords with the Boolean operative 'OR'
  • Select 'All Subheadings' if presented with the option
  • Free-text searches - for terms too new or not widely used enough to be Subject Headings
  • Search other databases
  • Use the "broad, sensitive" search checkbox with PubMed Clinical Queries

If too many hits:
  • Use Boolean operators - particularly 'AND' or 'NOT'.
  • Use specific Subheadings
  • Use filters - this will be done by using the Clinical Queries tool on PubMed to restrict to randomised controlled trials
  • Focus - available on many databases to focus on Subject Heading terms, retrieving articles with the Subject Heading as a main subject.
  • Use thesaurus terms
  • Use the "narrow, specific" search checkbox with PubMed Clinical Queries


Note: the goal of evidence based medicine is to elucidate the evidence behind interventions. Arbitrarily excluding papers due to a large volume of research risks excluding key findings

If too many irrelevant hits:

  • Use Subject Headings rather than free-text alternatives
  • Use the subject tree (see MeSH) to find more precise Subject Headings
  • Boolean operator 'NOT' - use with caution!

For thoroughness:

  • Avoid limits
  • Search reference lists
  • Specify inclusion and exclusion criteria
  • Include ALL relevant data
  • Should be reproducible

What if no relevant papers can be found?

Having searched the Cochrane Library,  PubMed Clinical Queries and other databases/grey literature mentioned above, for systematic reviews and randomised controlled trials/ cohort studies/case control studies etc,  the final stage would be to include a good quality review. This only really refers to a small subset of questions not easily answered by the other types of studies.

It may simply be that the current recommendations are based on institutional traditions, hence the importance of evidence-based medicine and its documentation..

With thanks to Marshall Dozier and the University of Edinburgh Library Staff for their advice on literature searching. The Edinburgh Medical Library website is very useful for tips on searching, and they have produced a useful pdf paper on how to systematically search the literature.


» Part 2: Accessing relevant articles

» Back to reviewer's toolkit