++
The most important part of a literature search is deciding what terms to enter as keywords. All electronic databases will utilize some form of algorithm to generate a list of citations based on the keywords you choose. Citations will generally be brought up if the keyword is present in the title or abstract of the article. Journal articles may also have lists of keywords assigned to them by the author that are then used to index the document. Because so many words have synonyms or alternative terminology, this process can be daunting at times. It is useful to begin by brainstorming the various words that may be useful in finding articles related to your topic. Consider the pathology or diagnosis, the population, the specific treatments or measurement tools, and the outcomes of interest.
++
Suppose we were interested in studying physical endurance in patients who have experienced a stroke, with specific interest in the effectiveness of exercise programs. We would want to learn about the previous success of exercise programs for this population, and how endurance has been measured. Some possible keywords might include:
++
++
Once we choose our keywords, we must consider how to combine them to generate citations that will relate to our question.
++
When performing a search, it is important to fine tune the choice of key words or reference terms to narrow the search. Most databases and search engines use a system called Boolean logic, named for George Boole, an English mathematician who invented it in the mid 1800s. Boolean logic utilizes three primary operators: AND, NOT and OR.∗
++
We will illustrate the use of these operators to develop a search strategy for our question related to endurance after stroke (see Figure 31.1).
++
++
AND narrows the search, requiring that all requested terms to be present. We can begin by finding references that include both of our main terms:
++
++
This search yields 6,423 citations.†
++
OR broadens the search, usually when you want to include synonyms of the main search term. We may want to expand the search so that it includes studies that use the terms stroke or cerebrovascular accident:
++
(stroke OR "cerebrovascular accident") AND exercise
++
This has broadened the search to 6,448 citations. The term in quotes is read as a fixed string, not individual words. The terms that appear in parentheses are searched first and then paired with the second command. If more than one set of parentheses is present, the search sequence goes from left to right.
++
NOT excludes concepts from the search. Suppose, for example, we want to eliminate studies that exclude patients with diabetes:
++
stroke AND exercise NOT diabetes
++
This brings the search down to 6,084 citations.
++
We may finally want to limit our search to outcomes related to endurance:
++
stroke AND exercise AND endurance
++
The search is now limited to 439 references. Perhaps we can refine it a little more by making our terms more specific. For example, let's specify "physical endurance" to clarify the type of endurance we want to include:
++
stroke AND exercise AND "physical endurance"
++
Now our search is down to 278 citations. This is still too large a number of references, however. Let's examine how we can move through this process more effectively.
+
++
++
++
The National Library of Medicine has developed a sophisticated system of Medical Subject Headings (MeSH) that have been adopted by most databases that cite medical and health literature. Because authors will use different terminology for the same concepts, MeSH terms provide a consistent search vocabulary. MeSH consists of sets of terms in a hierarchical structure or "tree" that permits searching at various levels of specificity.2 The tree starts with broad categories such as "Anatomy," "Disease" or "Analytical, Diagnostic and Therapeutic Techniques and Equipment." More specific terms, close to 23,000 descriptors, are provided at various levels of the tree. Each heading can be further detailed in a list of subheadings. Table 31.2 shows a partial list of terms to illustrate this hierarchy.
++
++
Additional options include exploding the search, which means that the search will retrieve results using the selected term and all of the more specific terms in the tree.
++
Focusing or restricting the search means that it will be limited to documents where the subject heading is the major point of the article.
++
OVID provides a "Map Term to Subject" feature that shows MeSH headings for entered keywords. PubMed provides a MeSH database that will serve the same purpose. These programs provide a useful tutorial for those who are not familiar with this process.
++
For our current example, the term stroke is not a MeSH heading, but "cerebrovascular accident" is. There are several terms that relate to exercise, including "exercise therapy," "exercise test," "exercise movement techniques," and "exercise tolerance." The definition of "exercise therapy" is closest to our intention: "Motion of the body or its parts to relieve symptoms or to improve function, leading to PHYSICAL FITNESS, but not PHYSICAL EDUCATION AND TRAINING." MeSH terms related to endurance include "physical endurance," "sports," and "muscle fatigue." The definition for "physical endurance" is most appropriate for our purpose: The time span between the beginning of physical activity by an individual and the termination because of exhaustion.
++
In our search we use a slash to define our terms as MeSH headings. By combining the three terms:
++
cerebrovascular accident/AND exercise therapy/AND physical endurance/
++
We now obtain 9 references, substantially different from the 278 we had before!
++
A comprehensive search strategy for this topic is illustrated in Box 31.1, showing the scope of this process. This search was used to generate references for a systematic review.3
++
Search strategies can be refined by setting specific limits or filters, including language, dates of publication, or studies that are provided as full text. Searches can be restricted to studies of humans or animals, specific age groups or gender, or types of studies such as case studies, randomized trials, or systematic reviews. A specific author or journal can be searched. When a search yields large numbers of potential references, such limits will often provide a useful way to hone in on articles that are available and contemporary.
++
PubMed has developed predesigned strategies to target a search for studies on therapy (interventions), diagnosis, etiology or prognosis. These search strategies can be accessed under Clinical Queries.
+++
Truncation and Wildcards
++
Most databases will recognize truncation and wildcard symbols that can be used to allow a broad interpretation of a search term, finding plurals, spelling variations and alternate forms of words. Each database specifies the symbol to be used, such as a question mark (?), an asterisk (∗) or a dollar sign ($). For example, entering the term "sensitiv$" in PubMed will retrieve articles that include "sensitive" or "sensitivity." The term "wom$n" will find "woman" or "women."
++
BOX 31.1 Search Strategy for Studies Related to Stroke, Exercise and Endurance
Sophisticated strategies that combine MeSH headings and keywords can provide a comprehensive search. Synonyms for keywords are incorporated to target the full scope of studies related to the topic. Special abbreviations can be used to clarify where terms are found:
exp = explode
ab = words in abstract
hw = word in subject heading
pt = publication type
ti = words in title
tw = text words
/ = MeSH subject heading
$ = truncation symbol
adj4 = adjacent (within four words)
or/ = apply OR to specified steps
The following search illustrates this process related to the effect of exercise on endurance following stroke.3 The strategy illustrates how MeSH terms, keywords and series of search groupings can be combined to retrieve references.
exp Cerebrovascular accident/
(stroke or cva$ or cerebrovascular accident or cerebral vascular).tw
exp Brain injuries/
exp Hemiplegia/
(hemipleg$ or hemipar$ or brain injur$).tw
or/1-5
exp Exercise Therapy/ or exp Exercise/
exp Physical Fitness/
exp Physical Endurance/
treadmill.tw
((aerobic or endurance or cardio$ or fitness) adj5 (train$ or program$ or protocol$ or intervention$)).tw.
or/7-11
exp Randomized Controlled Trials/
Clinical trial.pt
exp Random Allocation/
Random$.tw.
exp Cross-Over Studies/
Control$.tw.
Experimental$.tw.
exp Follow-Up Studies/
or/13-20
6 and 12 and 21
Adapted from Pang MY, Eng JJ, Dawson AS, Gylfadottir S. The use of aerobic exercise training in improving aerobic capacity in individuals with stroke: A metaanalysis. Clin Rehabil 2006;20(2):97–111.
+++
Broadening the Search
++
Several databases offer an option to search "related articles" or "similar pages," once a relevant citation has been found. By using this function, one relevant article becomes the key to a list of useful references. These references will usually be shown in order of their relevance to the original record.
+++
Science Citation Index
++
Science Citation Index is a reference system that allows researchers to search based on citations (see Table 31.1), by tracking literature forward or backward, and across disciplinary boundaries. The index will identify articles, books or other sources that have cited a particular reference or author. This index is available through online subscription or DVD, usually through a library.
+++
Other Articles' Reference Lists
++
The reference list at the end of published articles will provide a handy inventory of sources that were used by other authors in their review of literature. The obvious disadvantage of this approach is that you will only identify references that are older than the original article. But it is often a good place to start, and to distinguish classic references that are used repeatedly.
+++
Sensitivity and Specificity of Literature Searches
++
Electronic searching can be considered a blessing or a curse by the researcher. It is certainly more efficient than the old-fashioned time-consuming hand searching through volumes of the Index Medicus, although at the same time it can generate substantial lists of citations that may not be relevant to the project. The incentive for developing competence in searching is the ability to avoid retrieving too many citations or long lists of irrelevant ones.
++
In this aspect of searching, the terms "sensitivity" and "specificity" have been applied, with meanings analogous to their use with diagnostic tests (see Chapter 27).4 Sensitivity, also called recall, is the proportion of retrieved citations that are relevant, or the likelihood of finding relevant references. A sensitive search might include some irrelevant studies but is more likely to be comprehensive. Specificity, also called precision, is the proportion of citations that are relevant that the search is able to retrieve, or the likelihood of excluding irrelevant references. A specific search might exclude some relevant studies but is more likely to include the very relevant ones.
++
When a search results in too many citations, you need to improve specificity by narrowing the question, using more specific search terms, combining search terms, or setting limits. Using the "focus" option can also restrict the search. When the number of references is too small, the search needs to improve its sensitivity by broadening the question, finding better search terms, using truncation or wildcards or trying varied combinations of terms. Using the "explode" option can improve the breadth of the search. In PubMed, under Clinical Queries, choosing a "sensitive" search will explode the search to include a broad range of articles, some that may be less relevant. Choosing a "specific" search will focus the search on the most relevant articles, but will eliminate some that have less obvious relevance. A useful guide to terms that can be used to increase sensitivity or specificity has been developed for use with MEDLINE.5