PubMed Search Builder

Purpose

This skill drafts Boolean search strings for literature databases from a structured research question. A review can only be as complete as its search: the search strategy determines which records the review team ever sees, so small syntax errors or missing synonyms can change the conclusions of a review.

When to use this skill

• You are drafting a search strategy for a systematic review protocol.
• You are scoping the literature for a research proposal or thesis.
• You are building a search for a clinical question and want a readable starting point.
• You need database-specific syntax variants (Embase, CENTRAL, Web of Science) for a multi-database review.

When not to use this skill

• You need a final, runnable search for a registered systematic review without librarian review. This skill drafts; it does not replace PRESS peer review.
• You are searching for an individual patient. This is not a clinical tool.

How a search strategy is structured

A systematic review search strategy is built from concept blocks, one per PICO element. Each block contains the synonyms for one concept, joined by OR. The blocks are joined by AND. For example:

(Population synonyms)
AND
(Intervention synonyms)
AND
(Outcome synonyms)
AND
(Design filter)

The comparator concept is often omitted from the search because including comparator terms tends to reduce sensitivity, and most reviews want to capture records that compare the intervention with anything. This is a decision the review team makes, not a default this skill enforces.

Database syntax notes

PubMed

• MeSH headings use [MeSH Terms]; text words use [Title/Abstract].
• Publication types use [Publication Type].
• Truncation with * is supported (for example, "anticoagulant*").
• Phrase searching with double quotes is supported.
• The # character can be used to number lines, which makes strategies readable: #1 "Diabetes Mellitus, Type 2"[MeSH Terms].

Embase

• Emtree headings use /exp for exploded terms (for example, 'atrial fibrillation'/exp).
• Title/abstract uses .ti,ab.
• Publication types use .pt.
• Embase has a richer set of drug and device terms than PubMed, and its MeSH mapping is incomplete, so a review searched only on PubMed will miss records found in Embase.

CENTRAL

• CENTRAL uses MeSH and free-text terms. Field tags are less important; a simple combination of MeSH and truncated text words works well.

Web of Science

• Topic search uses TS=(), title uses TI=().
• No MeSH or Emtree thesaurus: only free-text terms, so synonyms matter more.
• Truncation uses * and wildcards vary by edition.

The PRESS checklist in practice

The PRESS guideline (2022 update) asks peer reviewers of search strategies to check: translation of the research question, Boolean and proximity operators, subject headings, text words, spelling and syntax, and filters. This skill builds its output so a PRESS review is feasible:

1. Concept blocks are kept separate and labeled.
2. MeSH and text word lines are distinct.
3. Terms that could not be verified are listed in review_notes.
4. Filters (study design, language, date) are flagged as requiring confirmation.

MeSH and Emtree verification

MeSH headings change with each annual update, and some concepts have no exact heading. This skill never invents a heading: if it cannot verify a heading, it lists it in review_notes. In practice you should check candidate headings against the MeSH browser (https://www.ncbi.nlm.nih.gov/mesh) and the Emtree browser in the Embase interface before finalizing.

Common failure modes

• Fabricated headings: inventing a MeSH term that does not exist. This is the most dangerous failure for a search tool. The review_notes field exists to catch it.
• Mixed syntax: using Embase syntax in a PubMed string. Each database has its own rules.
• Overly narrow searches: building a block with only one or two synonyms. Systematic review searches are intentionally sensitive; the goal is to retrieve, then screen.
• Missing truncation: failing to include plural and spelling variants.
• Skipping the comparator decision: leaving the comparator block in without the review team having decided to include it.

Integration with other skills

The output of this skill feeds the Abstract Screening Assistant and the Data Extraction Table Generator. The Evidence Synthesis Pack chains Research Question Builder, PICO Extractor, PubMed Search Builder, and Risk of Bias Assistant into one workflow.