81 lines
7.7 KiB
Markdown
81 lines
7.7 KiB
Markdown
# Turning a Research Question Into a Search Strategy That Finds the Right Papers
|
|
|
|
A literature search is only as good as the strategy behind it. Type a few keywords into a database and you will get results, but you will not know what you missed. A structured search strategy, built deliberately from the research question, tested and documented, gives you confidence that you have found the relevant evidence and lets others reproduce your work. This article walks through the process of turning a question into a search strategy, from concepts to final strings, and explains how [Yila](https://yila.ai), a source-linked research assistant, can help you understand and verify what the search returns.
|
|
|
|
## Start with a well-formed question
|
|
|
|
A search strategy inherits every ambiguity in the question it comes from. Before touching a database, write the question in a structured form. For intervention questions, population, intervention, comparison and outcome is the familiar framework. For questions about experiences, population, phenomenon of interest and context often fits better. For diagnostic, prognostic or exposure questions, adapted frameworks exist.
|
|
|
|
The framework is not the search. Its value is in forcing you to name each element precisely, so you can decide which elements to search for and which to apply later during screening.
|
|
|
|
## Choose the concepts to search
|
|
|
|
Not every element of the question belongs in the search. Searching for every element narrows results sharply and increases the risk of missing relevant studies, because titles and abstracts often omit some elements. A common approach is to search for two or three core concepts, typically the population and intervention or exposure, and apply outcome and design criteria at the screening stage.
|
|
|
|
Outcomes in particular are often poorly described in abstracts. Including them in the search can exclude relevant studies that report the outcome only in the full text.
|
|
|
|
## Build vocabulary for each concept
|
|
|
|
For each concept, build two sets of terms: controlled vocabulary and free text.
|
|
|
|
Controlled vocabulary consists of the subject headings assigned by database indexers. They group articles on a topic regardless of the words authors used. Find the appropriate headings by browsing the database's thesaurus and by looking at how known relevant articles have been indexed. Decide whether to include narrower headings beneath each term.
|
|
|
|
Free-text terms capture articles not yet indexed and concepts that the controlled vocabulary handles poorly. List synonyms, related terms, abbreviations, spelling variants and older terminology. Use truncation for word endings and wildcards for spelling variants, checking the specific syntax of each database. Use proximity operators where available to keep related words close together without requiring an exact phrase.
|
|
|
|
## Combine terms logically
|
|
|
|
Within each concept, combine terms with OR to capture any of them. Between concepts, combine with AND so that results must address all core concepts. Use NOT sparingly, if at all, because it can exclude relevant records in unexpected ways.
|
|
|
|
Write the strategy in numbered lines so that each concept is built separately and combined at the end. This structure makes the strategy easy to read, check and modify.
|
|
|
|
## Test the strategy against known studies
|
|
|
|
Before running the final search, test it. Collect a set of known relevant studies, perhaps from a previous review, key papers you already know or citations in recent articles. Run the strategy and check whether it retrieves them. If any are missing, examine how they are indexed and described, and adjust the terms.
|
|
|
|
This test set is one of the most effective ways to improve sensitivity. It reveals missing synonyms, overly restrictive combinations and indexing quirks that are otherwise invisible.
|
|
|
|
## Balance sensitivity and precision
|
|
|
|
A highly sensitive search finds nearly all relevant studies but returns many irrelevant ones. A highly precise search returns mostly relevant studies but misses some. Systematic reviews generally favour sensitivity, accepting a larger screening load to reduce the risk of missing evidence. Scoping exercises or rapid reviews may accept lower sensitivity for speed.
|
|
|
|
Validated search filters for study designs, such as randomised trials, can improve precision without sacrificing much sensitivity. Use published, tested filters rather than inventing your own.
|
|
|
|
## Translate across databases
|
|
|
|
Each database has its own controlled vocabulary and syntax. A strategy built in one must be translated, not copied, for others. Map subject headings to their equivalents, adjust truncation and proximity syntax, and check field codes. Test each translated version against your known studies.
|
|
|
|
Search several databases appropriate to the topic. No single database covers everything, and overlap between them is smaller than many people assume.
|
|
|
|
## Supplement database searching
|
|
|
|
Database searches miss some relevant work. Supplement them with citation searching, both backwards through reference lists and forwards through citing articles, searches of trial registries and grey literature sources, and contact with experts where appropriate.
|
|
|
|
## Understand what you have found
|
|
|
|
Once results are in, you need to understand them quickly. Loading key papers into [Yila](https://yila.ai) lets you ask how they define important terms, which populations they studied and what they measured, with answers linked to the passages that support them. This can reveal vocabulary you missed, prompting a refinement of the strategy, and helps you confirm that apparently relevant results really do address your question before investing in full screening.
|
|
|
|
## Work with a librarian when you can
|
|
|
|
Information specialists and academic librarians build search strategies every day and know the quirks of each database far better than most researchers. Involving one early, ideally when the question is being finalised, usually produces a more sensitive and more efficient strategy. Many institutions offer this support free to staff and students. Even a single consultation to review a draft strategy can catch missing headings, syntax errors and unnecessary restrictions. Peer review of search strategies by a second information specialist, using a structured checklist, is also increasingly expected for systematic reviews and is worth planning for.
|
|
|
|
## Keep the search up to date
|
|
|
|
Long projects need update searches before submission. Rerun each strategy with a date limit covering the period since the original search, screen the new results using the same criteria and record the update separately. If the strategy has changed, note how and why, so readers can understand any differences between the original and updated results.
|
|
|
|
## Document everything
|
|
|
|
Record the full strategy for each database, the date searched, the database platform and coverage dates, any limits applied and the number of results. Reporting guidelines require this, and it allows others to reproduce or update your search. Keep copies of the strategies with the project files.
|
|
|
|
## Common mistakes
|
|
|
|
- Searching for every element of the question.
|
|
- Relying on free text or controlled vocabulary alone.
|
|
- Not testing against known relevant studies.
|
|
- Copying strategies between databases without translation.
|
|
- Applying language or date limits without justification.
|
|
- Failing to document the search fully.
|
|
|
|
## Conclusion
|
|
|
|
A good search strategy is built, not typed. Start from a structured question, choose core concepts, develop controlled and free-text vocabulary, combine terms logically, test against known studies, translate carefully across databases, supplement with other methods and document everything. The result is a search you can trust and defend, and a body of evidence that genuinely reflects what is known about your question.
|