Before fleshing out your keywords, decide what databases you will search. You want to cast a wide net across several databases to ensure that  you pull back all (or almost all) relevant articles.

• One great strategy is to pull up recent systematic reviews on topics similar to yours and ones in journals you might choose to publish in. Tally which databases are most commonly searched and use this to inform your own selections.

  • While you are at it, look for sample search queries in these systematic reviews.

  • Take note of what limiters, if any, are used. Determine what record fields (e.g., title, abstract, index/subject terms) are typically searched. Use this information to make decisions about your own search queries later in the process.

• The following resources offer comparisons of databases and may help you determine which databases will best fit your systematic review. 

  • This article reviews the capabilities of various search engines and databases. 

  • This Michigan State LibGuide describes how Google Scholar differs from human-curated databases like MEDLINE.

• You can also look through subject guides to see which databases are recommended for a given discipline by subject librarians.

An A to Z list of all BYU database subscriptions can be found at https://lib.byu.edu/databases/.

The following table lists several databases commonly used in social and health sciences systematic reviews. The names on the left link to the database through BYU. The links on the right take you to various how-to guides. Databases share many features, but they also have unique elements such as their own search algorithms, search functions, and controlled vocabulary. Understanding these unique elements up front and how they work will save you time later.

A couple of things to note

• At BYU, CINAHL, APA PsycINFO, and MEDLINE can be searched on the EBSCO platform, which means that you could search all three databases simultaneously. On the plus side, this saves time. On the downside, deduplicating results is trickier and you may have to adjust use of controlled vocabulary because they may differ between databases.

• Most of these databases focus on peer-reviewed literature. If you are concerned about evidence selection bias (e.g., not pulling all relevant data because significant findings are more likely to be published than non-significant findings), you may also need to search gray literature and unpublished studies. See Paez (2017) for a viewpoint on why grey literature is important and how to use it in systematic reviews. See this King's College LibGuide about finding grey literature. 

Reference

Paez, A. (2017) Gray literature: An important resource in systematic reviews. J Evid Based Med. 2017; 10: 233– 240. https://doi.orr/10.1111/jebm.12266

Developing your sets of keywords will be an iterative process. One way to keep track of your keywords is to create a simple table like the following:

Each time you identify an additional synonym for a concept, add it to the appropriate set of keywords. Notice that each set is being built with "or" separating each word. This makes it easier to translate it into the database query later. In this example, multiple-word phrases are encased with double quotation marks. This is a common way to make sure databases search for full phrases rather than individual words. You might also decide to truncate words that would have multiple endings. This is often signified with an asterisk (e.g., injur* would retrieve injury, injuries, etc.). Usually you can use both truncation and quotations in the same phrase. However, because databases have their own search functions and algorithms, it is best to examine what works for each one.  

Additional tips for identifying keywords:

• Lead researchers on the project generally know the literature. Have them list keywords for each concept.

• Run a search for each concept individually in one or more databases. Look for articles relevant to your concept. Then, pull new keyword variations from titles, abstracts, author keywords, and subject lists. 

• Find the list of controlled vocabulary for a database and search for each of your concepts. See what related terms are listed and pull those that are relevant. 

• If relevant, use the acronyms PICO (Population, Intervention, Comparison, Outcome) or SPIDER (Sample, Phenomenon of Interest, Design, Evaluation, Research type) to develop sets of keywords. See Cook et al. (2014) for a comparison of using these approaches for a qualitative evidence synthesis. 

The scope of your review and the popularity of your topic will impact how many initial results you get from all of your database searches. Sometimes you get thousands. Sometimes you get hundreds. When possible, using database limiters can be an easy way of applying exclusion criteria from the beginning. Some potential limiters are the following:

• Publication date range (e.g., if you are updating a previous review)

• Language (e.g., if your research team will only be able to read articles in certain languages)

• Source type (e.g., empirical, peer-reviewed)

• Species

• Sex

As you think about applying limiters, consider whether they will introduce bias into your process. For example, if you limit to only peer-reviewed articles, you may exclude relevant theses and dissertations. Also keep in mind that not all databases have the same limiters.

Also, depending on how you are tracking your process, you should note the number of results in each database before and after you apply limiters. 

Remember to document exactly how you do your searches so you can convey the process clearly in your methods. The goal is replicability. Ideally, you want to run searches as similarly as possible across databases. However, your search may be adapted somewhat to accommodate the search functions and controlled vocabulary of each database. 

The following table illustrates how to translate a search across several databases.

One Thing to Note

Most databases allow you to combine searches. When using this method, you would run a separate search for each set of keywords (e.g., search #1 for brain injury, search #2 for working memory) and then use an advanced search function to combine the searches (e.g., #1 AND #2). Some may use this to exclude a body of research. For example, you could run a search for the topic sport, which would become search #3, and then do a combined search with the first 2 (e.g., #1 AND #2 NOT #3).

Time-Saving Tip 

Before downloading or exporting the results from all database queries, it is useful to have a couple of team members assess whether they seem appropriate.

• Look through the first page or so of results. Are potentially relevant and expected articles showing up?

• Look at the sum total of results from all queries. Does the sum seem reasonable based on your knowledge of the literature or does the search need to narrow or broaden?

After running your finalized query in each database, you will need to download the records, combine them, and deduplicate them. You can use bibliography mangers, programs like Covidence, or a spreadsheet. The following table describes how to download records to an Excel spreadsheet. 

Tracking record numbers is important when it comes time to write the methods and results of your systematic review. Many publications include a figure like the PRISMA flow chart that tracks the number of articles originally found, the number after deduplicating, the number that met inclusion criteria, etc. The PRISMA checklist is also helpful when drafting content for your methods/results. The Record Keeping page in this guide addresses additional areas where you will need to track details of the research process.