acebcp

A Search Strategy is a plan. You will find it easier if you have spent time framing your question in terms of:

Is St Johns Wort more effective than traditional antidepressants in the treatment of depression?

P = depressed people
I = St Johns Wort
C = antidepressants
O = reduction of depressive symptons

Plan how you will combine your searches. This will vary slightly for each database, but basically requires you to decide which concepts you wish to combine. Generally you would start with a broad search, and then narrow it down. Typically this would mean combining the Population and Intervention concepts first.

Using the example and a fictional database, the first concept, population, is depressed people. Identify a term likely to be used in common language for this, eg, depression.

Set
1

Search Term
Depression

Hits
10,000

The next concept is the intervention, St Johns Wort. Databases usually have specific rules for searching phrases, but ignore this for this example.

Set
1
2

Search Term
Depression
St Johns Wort

Hits
10,000
2,000

You really want all records that include both concepts. To do this, combine them using the 'AND' operator.

Set
1
2
3

Search Term
Depression
St Johns Wort
1 and 2

Hits
10,000
2,000
1,300

You might find 1300 records, ie, 1300 records contain both the 'population' and the intervention concepts. If you have a specific comparison, such as ' antidepressants' , you could then 'and' this. By doing this you are narrowing your search. This might reduce your search down to 100. In other words, 100 records share all three concepts of Population, Intervention and Comparison.

Set
1
2
3
4
5

Search Term
Depression
St Johns Wort
1 and 2
Antidepressants
3 and 4

Hits
10,000
2,000
1,300
30,000
100

In the example, we have used 'natural language' to describe our concepts. Databases use both natural language and controlled language subject terms. Natural language is sometimes referred to as textword or keyword searching. Most Internet sites use natural language, and most bibliographic databases allow you to use both.

One of the drawbacks of textwords, is that you need to be able to identify possible alternative expressions for the same concepts. Using our example, what if there was an excellent systematic review on the topic, but it didn't use the term 'St Johns Wort'. What if it were called 'a systematic review of hypericum in patients with mild to moderate depression'?.

There are two strategies that you can use. Firstly, identify alternative words to describe the concepts and combine them with an 'OR' operator. Looking at our example,

Set
1
2
3
4
5
6
7

Search Term
Depression
St Johns Wort
Hypericum
2 or 3
1 and 4
Antidepressant$
3 and 4

Hits
10,000
2,000
3,000
4,000
1,600
35,000
150

Without using the word 'hypericum' we retrieved 100 references. By including it, we broadened our search strategy to retrieve 150 records. Set 4 does not equal 2,000+3000, because some of these records contain both 'hypericum' and 'St Johns Wort'.

The second strategy is that of truncation. Notice how Set 6 contains the symbol '$' . It is shorthand way of using the word 'or'. It actually means, search for 'antidepressant or antidepressants'. Hence it also broadens your search.

For example, child$ would search for 'child OR children' . The truncation rule is that you take the last common letter:

Truncation symbols allow you to search for plurals. They vary across databases. In OVID Medline a $ sign is used, but in PubMed and the Cochrane Library an * is used.

The use of the words AND and OR is commonplace in all databases and searching the Internet. It is called Boolean Logic. A handy way to remember this is that OR means more. It broadens your search. AND means less and narrows your search.

Some databases even have wildcards to use in the middle of words. For example, OVID Medline

The other type of searching is called 'controlled language' or 'subject searching'. The most commonly known one in medicine is called MESH. This is a list of thesaurus terms. When a database uses a controlled language such as MESH, it means that someone has read the articles indexed in the database, and given them a range of subject headings taken from the Thesaurus. You can then search the database using these. Databases such as Medline, Embase and Cinahl allow you to do textword searching and subject searching.

The best way to explain the difference, is by use of an example where the concept of interest is the 'hand' . If you search Medline using it as a textword, you will retrieve a large number of records. Many of these will focus on the 'hand' but many will be irrelevant because they will contain expressions such as 'on the other hand' or 'hand it over' .

If you search for it using a MESH heading, you will restrict your search to only those items that are about the 'hand'.

Subject searching is more specific and gives you more control. Subjects normally have a range of subheadings - making searching easier. For example, myocardial Infarction/rh will find everything on myocardial infarction and rehabilitation.

Textword searching is good for recent developments (where there may be no formal subject heading), or when something is so rare that you are looking for anything at all that might use the term. Experienced searchers use both to build up complex strategies.

Most databases have online help in providing you with the controlled vocabulary. It is sometimes called mapping. In OVID Medline for example, if you enter a natural language term such as stroke, it automatically directs you to the MESH heading "cerebrovascular accidents". Not all databases use the same language. Embase, Medline and CINAHL all use different controlled languages.

The next step in searching is to limit your results. There are some standard limits built into most databases, such as year and language. However an evidence based approach means we are really looking to narrow our search down to particular types of evidence, in particular, the highest quality ones, so we need to have some strategies for doing this. These are referred to as "quality filters" or "hedges". There are some quite complicated ones available via the Internet, but there are some simple choices you can make when using databases restrict your search results to higher quality evidence. When you use PubMed Clinical Queries, you are actually using such filters, which have been developed and saved by experienced searchers.

In terms of Boolean Logic, logically you are combining your final search results from a strategy (developed from PICO) and the quality filter.

Consider a search on Medline. You can restrict your search to a particular study type, eg, a RCT, clinical trial, or a Cochrane Review. Similarly on CINAHL, you can restrict to a systematic review, a clinical trial, or research.

If 'formal limits' are unavailable, you can always add in the words 'review' or 'systematic review' to your search to limit the findings.

Sensitivity is the likelihood of retrieving relevant items, whereas specificity is the likelihood of excluding irrelevant items.

Identifying the key concepts of interest (ie. PICO)
Identifying where you will search
Combining concepts with each other and applying appropriate limits

Typically start with Population and Intervention
Use OR to broaden your search
Use AND to combine your PICO concepts
If using textwords, identify all possible alternatives and consider truncation symbol for plurals, alternative spelling and all words with the same word stem
If you get few or no results, check your spelling!
Once you have completed your search, consider applying a filter (ie, this will AND your final search with a limit or search which will identify better quality evidence)
A database will only ever be as good as the information it indexes in the first place. There is no one database that will provide you with all answers. If you do find something, it isn' t necessarily good evidence. If you don' t find anything, it doesn' t mean it doesn' t exist.

Resources

How to Construct a Search Strategy