How to Combine Keywords and Subject Headings in a Search Strategy

A search strategy is the part of a paper you can actually control when the deadline is close. Before you write a single sentence of your literature review, you decide which concepts matter, which words describe them, and how those words get combined into queries you can run, record, and defend. Cochrane's guidance on searching for studies starts from eligibility criteria and builds outward: plan the search, combine free-text terms with subject headings, use multiple sources, document the process, deduplicate, screen in stages, and rerun before publication. That sequence is the backbone of this article.

This guide walks through a complete, repeatable workflow: separate your concepts, map natural language to controlled vocabulary, combine synonyms with OR, combine concepts with AND, and test the strategy against papers you already know should appear. Each stage has a diagnostic checkpoint so you know whether to continue or fix something first. It ends with a worked student example and a concrete submission decision.

The most common failure is beginning with words. Instead, begin with what makes a paper eligible for your review. Write down the population, the intervention or exposure, the outcome, and any study-design limits you intend to apply. Cochrane explicitly recommends planning searches from eligibility criteria, so treat that list as your source of truth.

Checkpoint 1: Read your eligibility criteria aloud. If you cannot point to the phrase in your criteria that justifies each concept you plan to search, drop that concept or rewrite the criteria. A concept with no criterion behind it will pull in noise you then have to screen out by hand.

A concept is a single idea your review depends on. A keyword is one way of expressing that idea. Keeping these separate prevents the classic mistake of stuffing every synonym into one long string and losing track of which idea failed to retrieve results.

Build a concept table with one column per concept. Under each column, list the natural-language words you would use, then the controlled vocabulary terms that describe the same idea. Leave the Boolean operators out for now; you are only organizing.

Checkpoint 2: Every column should be justifiable from your eligibility criteria, and no column should contain two unrelated ideas. If a column has grown into a paragraph, split it.

Controlled vocabulary is the fixed set of subject headings a database assigns to records. Free-text keywords match what authors wrote; subject headings match how indexers categorized the paper. Cochrane's recommendation to combine free-text terms with subject headings exists precisely because these two retrieval routes catch different papers.

PubMed's documentation is a useful model for how to do the mapping. It supports field tags, phrase searching, filters, Advanced Search, saved searches, Similar Articles, citation lookup, and links to full text when available. Crucially, PubMed's search details reveal how your terms were translated and combined. That translation view is your mapping tool: type a natural-language term, then read how the database interpreted it before you commit.

Checkpoint 3: For each concept, you should be able to name at least one free-text term and, where the database offers one, at least one subject heading. If a concept has no subject heading, note that explicitly rather than inventing one.

OR broadens. Within a single concept column, join every synonym and every subject heading with OR so that any one of them can retrieve a record. This is the step that protects you from missing papers that use a different word for the same idea.

A practical pattern for one concept looks like this in structure (not in any specific database's syntax):

• (free-text term 1 OR free-text term 2 OR subject heading for the concept)

Keep each concept's OR block in its own parentheses. That way, when you combine concepts later, the grouping stays intact and you can test one block at a time.

Checkpoint 4: Run each OR block on its own and note roughly how many records it returns. A block that returns almost nothing usually means a spelling or truncation problem; a block that returns an implausibly large set usually means a term is too generic.

AND narrows. Once each concept block retrieves sensibly on its own, join the blocks with AND so a record must satisfy every concept to survive. The structure becomes:

• (Concept A block) AND (Concept B block) AND (Concept C block)

This is where a strategy either becomes precise or collapses. If your AND combination returns zero results, the problem is almost always one over-narrow block, not the AND itself. Return to Checkpoint 4, loosen the offending block, and recombine.

Checkpoint 5: Compare the record count of the full AND combination against each individual block. The combined set should be smaller than any single block. If it is not, your parentheses or operators are wrong.

A search strategy is only as good as its ability to find the papers you already trust. Before you screen anything, assemble a small set of known relevant papers — ones your supervisor cited, ones in your course reading, ones that clearly belong in the review. Then check whether your strategy retrieves them.

This is the missed-known-paper test. For each known paper that fails to appear, diagnose why: Is the concept missing from your table? Is the author's term absent from your OR block? Does the paper use a synonym you did not include? Does it lack the subject heading you relied on? Each failure is a specific, fixable gap, and fixing it usually improves recall for papers you have not yet found.

Checkpoint 6: Every known relevant paper should be retrievable by your strategy, or you should be able to explain exactly which concept or term is responsible for the miss and add it. Repeat until the known set is covered.

Tooling helps most when it is attached to a specific step rather than sprinkled across the workflow.

• Finding candidate papers: Run your drafted queries through a scholarly search interface such as WisPaper search, and read the public explanation of WisPaper Scholar Search if you want an entry point for finding academic papers.
• Organizing what you retrieve: As records accumulate, save or upload the papers you intend to keep and organize them in My Library. This keeps your known-paper test set and your screening set in one place.
• Interrogating your saved set: When you need to ask questions based on papers you have uploaded to My Library or added from search results, use Library QA.
• Checking references before submission: Before your bibliography goes into the paper, run your BibTeX through TrueCite, which checks BibTeX against real academic databases to flag hallucinated references.
• Longer synthesis tasks: If your task grows beyond retrieval into a literature review or reproduction effort, the Scholar Agent area covers literature review, inspiration, and reproduction use cases.

Suppose you are writing a course paper on a specific intervention for a specific population, the rubric rewards a documented search strategy, and you have limited database access. Your supervisor's feedback on the proposal said your scope was too broad.

1. Eligibility criteria. Narrow the population and the intervention until each fits one sentence. The supervisor's feedback is your justification for cutting a third concept entirely.
2. Concept table. Two columns: population and intervention. Under each, list the words you would naturally use, then the subject headings the database offers.
3. OR blocks. Combine synonyms and headings inside each column. Run each block alone and note the counts.
4. AND combination. Join the two blocks. If the count is zero, loosen the intervention block first, since it is usually the more specific one.
5. Known-paper test. Take three papers your supervisor cited. If one is missing, check whether it uses a synonym absent from your intervention block, add it, and rerun.
6. Document. Record the concepts, terms, headings, operators, and the date you ran each query, because Cochrane recommends documenting the process.
7. Screen in stages. Cochrane recommends staged screening and deduplication; remove duplicates before reading titles and abstracts.
8. Rerun before submission. Cochrane recommends rerunning searches before publication, so refresh your queries on the day you finalize the reference list.

Clear rules prevent you from polishing a search forever.

• Stop when every known relevant paper is retrieved and each concept block returns a plausible count on its own.
• Stop when adding another synonym no longer changes which known papers you retrieve.
• Escalate to your supervisor when a concept has no controlled vocabulary and your free-text block still misses known papers after two rounds of synonym expansion.
• Escalate when your eligibility criteria themselves are ambiguous — no amount of query tuning fixes a scope problem.
• Escalate when your combined search returns zero and loosening each block individually has not resolved it; the issue may be a database limitation rather than your terms.

Return to the student example. The rubric rewards a documented strategy, access is limited, and the supervisor flagged over-broad scope. The submission decision is not "is my search perfect" but "can I defend every choice in it." If your concept table traces to your criteria, your OR blocks retrieve their own concepts, your AND combination narrows correctly, and your known papers all appear, you have a defensible strategy. Document the queries and dates, deduplicate, screen in stages, rerun on finalization day, and submit. If any checkpoint still fails, fix that one checkpoint rather than restarting the whole strategy.