The development of a search strategy is an iterative process: one attempt will rarely produce the final strategy. Strategies are usually built up from a series of test searches and discussions of the results of those searches among the review team.
The first step is to break down the review question to help guide the development of search terms, using a structure such as PICOS.
Patients undergoing hip replacement
Any type of study design
It is not necessary to include all of the PICOS concepts in the search strategy. It is preferable to search for those concepts that can be clearly defined and translated into search terms. Concepts that are poorly defined, not likely to be included in journal abstracts, or not indexed in a consistent way, will be difficult to identify from database searches. If this is the case, using a broader search and then sifting through the identified studies may be preferable. This may apply particularly to the outcome(s) of studies as these are frequently not referred to in either the title or abstract of a database record.
Search filters are tested and in some cases validated strategies that can be used in a named database to identify specific types of study. They usually consist of a series of database index terms relating to study type combined with free text terms describing the methods used in conducting that type of research. There are filters available that will, for example, reliably identify RCTs in MEDLINE and in EMBASE, but filters for use in other databases or to identify other study types are limited. The development and validation of filters to identify other study types, such as diagnostic accuracy studies and qualitative research, is ongoing.1, 2, 3, 4 A useful source of information about search filters is the website maintained by the InterTASC Information Specialists’ Subgroup www.york.ac.uk/inst/crd/intertasc/ which lists both published search filters and research on their development and use.
Once the concepts of the search have been determined, the next stage is to produce a list of synonyms, abbreviations and spelling variants which may be used by authors. Similar research is often described using very different terms. To reflect this variation, a search strategy will usually comprise both indexing terms (if the database has a thesaurus or controlled vocabulary) and ‘free text’ terms and synonyms (from the database record’s title and abstract) to ensure that as many relevant papers are retrieved as possible. For example, when searching MEDLINE for studies about myocardial infarction, the free text term “heart attack” should be used as well as the Medical Subject Heading (MeSH) term “Myocardial Infarction”. Identifying appropriate indexing terms can be done by searching for key papers and checking how they have been indexed, consulting clinical experts in the review team and advisory group, as well as by scanning the thesaurus for relevant terms.
When selecting free text terms to use in the strategy it is important to take account of alternative spellings (including US and British English variants), abbreviations, synonyms, geographical variation, and changes in terminology over time. Sometimes it can also be useful to search for common mis-spellings, for example “asprin” when you want to retrieve studies of aspirin.
The important thing is to compile imaginatively and to check the indexing terms used in known relevant publications. Once a list of potential search terms has been compiled for each of the concepts, the next stage is to identify relevant subject headings which have been used to describe the topic in the databases you plan to search. For example with post-operative infection the following Medical Subject Headings (MeSH) are available for use in MEDLINE:
Surgical Wound Infection
Some of these terms are "high level" that encompass narrower or more specific terms. To capture these narrower terms, in those databases that offer the facility, it is possible to ‘explode’ the high level term and so search for many terms at once. The explosion facility within a database makes use of the hierarchical thesaurus. Using the command “exp Bacterial Infections/ ” in the OvidSP interface to MEDLINE will retrieve papers indexed with that term but will also automatically retrieve papers indexed with the narrower terms Bacteremia, Hemorrhagic Septicemia, Central Nervous System Bacterial Infections, etc. as displayed in the section of the MeSH below.
Central Nervous System Bacterial Infections
Eye Infections, Bacterial
The subject headings should be added to the concept list relating to the post-operative infection concept so that a first test search strategy for MEDLINE includes a mixture of text terms and MeSH headings.
bacterial adj infect$.ti,ab.
(postoperative adj complication$ or post adj operative adj complication$).ti,ab.
surgical adj wound adj infection$.ti,ab.
prosthesis-related adj infection$.ti,ab.
hip adj replacement adj3 infection$.ti,ab.
infection adj control.ti,ab.
bacterial adj contamination.ti,ab.
exp Bacterial Infections/
exp Postoperative Complications/
Surgical Wound Infection/
exp Infection Control/
The search has been written for the OvidSP search interface to MEDLINE and has commands specific to that interface:
adj Words have to appear next to each other. Also retrieves hyphenated words.
adj3 Words have to appear within 3 words of each other. Other numbers can be used as required.
$ Truncation symbol, for example ‘complication$’ retrieves 'complications' as well as 'complication'.
.ti,ab Restricts the search to title and abstract fields, to avoid retrieving unexpected results from the subject headings.
EXP Explode the subject heading, to retrieve more specific terms
/ MeSH heading.
? Optional wild card character used within, or at the end of, a search term to substitute for one or no characters. Useful for retrieving documents with British and American word variants.
Each database interface has its own unique set of commands and, information about these will be on the database help pages.
Once a series of concepts that reflect the PICOS elements have been compiled they are then combined using Boolean logic (AND, OR, NOT) to create a set of results which should contain articles relating to the topic in question. The AND operator is used to ensure that all the search terms must appear in the record, for example searching for “prostate AND cancer” retrieves all records which contain both the term prostate and the term cancer. AND is used to narrow down or focus a search.
OR is used to accumulate similar search terms and thus makes searches larger. Searching for “heparin OR warfarin” retrieves all records where either heparin or warfarin or both are found. It is best to use the OR operator to combine terms relating to the same concept (e.g. all the postoperative infection terms in the example above) before narrowing down a search using the AND operator with another set of terms.
NOT is used to exclude records from a search. For example, “acupuncture NOT asthma” will retrieve all records which contain the term acupuncture, but not those which also contain the word asthma. NOT should be used with great care because it may have a larger effect than anticipated; a record may well discuss both the concept of interest and the one to be excluded.
The combination of concepts using the Boolean operators might develop as follows (for MEDLINE using the OVIDSP interface):
1 Hip Joint/
2 Hip Prosthesis/
4 hip replacement$.ti,ab.
5 total-hip replacement$.ti,ab.
6 total joint replacement$.ti,ab.
7 hip surgery.ti,ab.
8 hip operation$.ti,ab.
9 (hip adj3 prosthe$).ti,ab.
10 (hip adj3 arthroplasty).ti,ab.
11 1 or 2 or 3 or 4 or 5 or 6 or 7 or 8 or 9 or 10
12 exp Bacterial Infections/
13 exp Postoperative Complications/
14 Surgical Wound Infection/
14 Prosthesis-Related Infections/
17 exp Anti-Infective Agents/
18 exp Infection Control/
19 exp Antibiotics/
20 Antibiotic Prophylaxis/
21 ((bacteri$ or wound$) adj2 (infect$ or contamin$)).ti,ab.
26 (anti$ adj infect$).ti,ab.
30 11 and 29
Sets 1 to 10 capture the concepts of hip replacement or hip surgery and are combined using OR to produce result set 11. Sets 12 to 28 capture the concepts of infection and infection prevention and are combined using OR to produce result set 29. The two sets of concepts are then combined to find the records which contain both concepts using AND to produce set 30.
The draft strategy can be tested on one database and the results checked by whether it retrieves papers that are already known to the team but were not used to develop the draft strategy. In addition, a small sample of the results of the test or scoping search can be examined by the review team to identify additional search terms (text words and indexing) or highlight potential limitations. The sample records need to be representative so bear in mind that the search results as output from the database will be listed in either alphabetical order by authors name, or by publication date or by date added to the database. Depending upon the complexity of the review topic, and consequently the search to be undertaken, this process may need to be repeated several times until an agreed strategy is formulated. If at all possible, the final search strategy should be peer reviewed to check for errors (spelling mistakes, incorrect use of operators, or failure to include relevant MeSH) that could reduce the recall of papers.5
At this point, the searches using other databases and resources can begin. However, this does not mean that search iterations should necessarily stop. If new search terms are identified during the review process they should be incorporated into the strategy or supplementary searches should be carried out.
Converting a final strategy for use in other databases requires care. While free text terms can usually be re-used in other databases you will need to identify one or possibly more matching relevant thesaurus terms used by the other databases. Each database thesaurus is unique so this procedure should be undertaken for each database being searched. For example, if you are searching MEDLINE for papers about “pressure sores” you would use the MeSH term “pressure ulcer” while if you were searching EMBASE you would need to use the EMTREE term “decubitus”.
If the search interface is also different you will need to make appropriate changes to the search operators used in the strategy. For example, some database providers use ‘$’ as the truncation symbol, while other database providers use ‘*’.
Not all databases include an abstract in the record and if this is the case you may want to make your search strategy more sensitive as you are relying solely upon terms being identified in the title (and any indexing fields). You can do this by using more synonyms and broader terms.
In some cases databases with web interfaces have a restricted range of search options and if this is the case searchers need to adopt pragmatic approaches and use very simple searches. For example, if there are limited options for combining terms using Boolean operators such as AND an alternative approach may be to run a number of separate searches on the database in place of one longer search.