Use PubMed to frame a clear question, build a precise MeSH-based search, screen with PRISMA steps, extract data, and report with full transparency.
What counts as a systematic review
A systematic review is a structured study of studies. It starts with a preplanned protocol, uses repeatable searches, screens records against explicit criteria, extracts data in a standard way, assesses bias, and presents findings with a flow diagram and checklist. The aim is to reduce random searching and selective reporting, so that readers can retrace the steps and reach the same set of included studies. PubMed is the most used gateway for biomedical literature, and with the right plan it can serve every stage from precise queries to clean exports.
How a PubMed review differs from a quick search
A quick search answers a narrow question for a single user. A systematic review answers a defined question for the field. That means an advance protocol, a full search string with synonyms and controlled vocabulary, saved search history, exports in standard formats, and a public report that shows every decision. The sections below show each step in PubMed.
| Step | What you produce | Helpful tools |
|---|---|---|
| Define the question | PICO elements and eligibility criteria | Protocol form; PROSPERO template |
| Write the protocol | Objectives, methods, outcomes, timeline | Registration at PROSPERO |
| Build the search | MeSH and keyword blocks, Boolean logic | PubMed Advanced, Search Details |
| Run and save | Search history, alerts, .nbib exports | My NCBI, Save Search |
| Deduplicate | One clean set of records | Citation manager, PRISMA flow counts |
| Screen | Title and abstract decisions by two reviewers | Screening tool; calibration set |
| Full text review | Include or exclude with reasons | Standard reasons list |
| Extract data | Aligned tables of study data | Shared sheet with codebook |
| Assess bias | Per study judgments and notes | Risk of bias checklist |
| Synthesize | Narrative or meta-analysis | Stats package; forest plot |
| Report | Text, tables, PRISMA 2020 checklist and flow | Checklist file and flow template |
Plan your question and protocol
State one primary question. Many teams use PICO: population, intervention or exposure, comparison, and outcomes. Write explicit inclusion and exclusion rules that map back to those elements. Add study designs you will accept, the date span, languages, and settings. Decide which outcomes are primary and which are secondary, and decide how to handle conference abstracts or preprints.
Turn the outline into a protocol. Post it before searching so that choices are time stamped and public. Health reviews often register at PROSPERO, a free registry that accepts health related topics. Registration helps avoid duplicate work and shows readers that the plan existed before results were known.
Build a PubMed search strategy
Good searches blend natural language with the PubMed controlled vocabulary. In PubMed, controlled vocabulary terms are Medical Subject Headings, or MeSH. Combine both so that you catch new articles not yet indexed and older records with nonstandard wording. Use the Advanced builder and the Search Details pane to check how PubMed is translating your words.
Turn concepts into MeSH and keywords
List the core concepts from the PICO. For each concept, note one or two MeSH terms plus a short list of free text synonyms. Consider plural and British or American spelling. Add a wildcard for endings when it helps. Keep each concept in its own set of parentheses, then join the sets with AND. This keeps the string readable and reduces missed combinations.
Use Boolean, fields, and filters with care
Within each concept block, join synonyms with OR. Use field tags to focus the meaning, such as [tiab] for title and abstract, or [mh] for subject headings. Put quotes around fixed phrases only when you need that exact wording. Filters can narrow results by study type, language, or date. Apply them only after you confirm that the base string is sensitive enough. The PubMed Help page explains filters and fields in depth, and you can keep it open in another tab while you build your search.
Test and refine in the Advanced builder
Open the Advanced page and add each concept to the query box with the OR and AND buttons rather than typing every operator. The builder prevents small syntax slips and logs each trial in your history. Review the Search Details line for automatic term mapping and fix any terms that map to the wrong concept. Save a versioned copy of the final string in your protocol and in a plain text file.
Example concept blocks
(“heart failure”[tiab] OR cardiomyopath*[tiab] OR “Heart Failure”[mh]) AND (telerehab*[tiab] OR “telemedicine”[tiab] OR “Telemedicine”[mh]) AND (exercise[tiab] OR “Exercise Therapy”[mh])
Doing a PubMed systematic review the right way
Once the search string is ready, rerun it without filters to capture the widest set, then add any date or language limits you planned in the protocol. Export the full set in .nbib format for use in a citation manager. Keep a copy of the raw export that matches the time stamp in your protocol. If you set an alert in My NCBI, rerun and export again before final data extraction to catch new records.
Export, deduplicate, and track counts
Use your citation manager to combine database exports and remove exact and near matches. Record how many records came from PubMed and how many were removed as duplicates. Those figures feed the PRISMA flow diagram later. Keep one clean library that moves forward to screening. Many tools can tag records with reviewer decisions and reasons, which keeps the audit trail easy to follow.
When exporting from PubMed, use the MEDLINE or PubMed format if your software can parse it, or select .nbib for broad compatibility. If you also search other databases, label each export by database and date in the file name. After deduplication, take a snapshot of counts per source and store it with the search log so that the flow diagram can be filled without guesswork.
Steps to perform a systematic review on PubMed
Screening happens in two rounds. First comes title and abstract screening against inclusion criteria. Two reviewers work independently and mark include, exclude, or unsure. Resolve differences by discussion or by a third person. Carry forward the set marked include plus the records marked unsure. The next round is full text screening where reasons for exclusion are recorded against a standard list.
Run dual screening without drift
Start with a small calibration set, such as fifty records, and discuss edge cases. Update the written rules with examples so that later decisions stay aligned. Randomize the order to reduce bias and blind the authors where possible. Track agreement rates and check a sample of excludes near the end to make sure criteria did not drift.
Extract data with a codebook
Build a shared sheet before reading full texts. Include citation fields, design, setting, sample, follow up, and the outcomes tied to the question. For interventions, include dose, delivery mode, and comparison. For observational studies, include exposure definitions and measures. Define every field in a short codebook so that different people record the same thing the same way.
Judge risk of bias
Pick one tool per design and stick with it. For randomized trials, many teams use a domain based tool with signaling questions. For cohort and case control studies, use a tool that covers selection, measurement, and confounding. Record judgments with short notes that point to page or figure numbers. Keep judgments independent, then reach consensus.
Write up methods and results
Readers need to see what you planned, what you did, and what you found. Methods should show databases, dates of last search, the exact PubMed string, any manual checks, screening rules, and data items. Results should show a flow diagram with counts at each stage, a table of included studies with key features, and a clear answer to the question. Use plain language to state what the studies agree on and where certainty is thin.
Follow PRISMA 2020 for reporting
PRISMA 2020 sets out items for transparent reporting of reviews. It covers the abstract, rationale, eligibility criteria, sources, search, selection, data items, risk of bias, synthesis, and more. The site hosts the checklist and the flow diagram templates you will need: PRISMA 2020. Filling the checklist as you draft saves time and reduces missed items.
Show the flow of records
Use a PRISMA 2020 flow diagram with counts for records identified, records after duplicates, screened, full texts assessed, excludes with reasons, and studies in the final set. The PubMed count in the first box should match the export log. If you reran searches before final extraction, note the date and any extra records found.
Common PubMed tags and smart use
Field tags and filters can sharpen retrieval when used with restraint. The table below lists tags many teams find handy and a one line tip for each. Use the Advanced page to add tags without typos and keep the Help page open for quick checks while you build and test strings.
| Tag or filter | Use | Tip |
|---|---|---|
| [tiab] | Search title and abstract text | Mix with phrases and roots |
| [mh] | Search MeSH headings | Add free text for in process records |
| [majr] | Major MeSH topic | Use when results are too broad |
| [pt] | Publication type | Pair with RCT, review, guideline |
| date filter | Limit by year range | Match the protocol span |
| language | Limit to chosen languages | State the reason in methods |
Tips for sensitive yet precise strings
Balance recall and precision. If results are sparse, widen concept wording, include both spellings, and remove limits not set in the protocol. If results are huge, add a key MeSH term, try [majr], or tighten the population. Read the first few pages of results to learn new synonyms and add them. Run a quick check that known benchmark studies are retrieved by your final string.
Watch for automatic term mapping surprises. PubMed may map a common word to a narrow concept or a gene name. The Search Details box shows the exact translation, so copy it into your log and edit the raw query if the mapping is off. Small tweaks, such as adding a field tag or a missing synonym, can lift recall without flooding the results list.
Keep an audit trail
Save the final string, the exact run date, and the number of hits. Store exports with file names that include the database and date. Capture a screenshot of the Search Details box for the record. Archive the protocol, the codebook, screening logs, and the PRISMA checklist files with a shared link inside the report.
When to seek librarian input
A trained librarian can turn vague terms into a strong string, spot missing synonyms, and suggest field tags that improve recall. Invite one early for a quick check of the plan and again after a pilot search so that refinements are based on records you saw. Many medical libraries post guides on PubMed searching that mirror the steps in this guide and offer contact forms for one to one help.
Ethics, equity, and access
Keep inclusion choices fair. Avoid language limits unless you can justify them. Document conflicts of interest for all reviewers and report funding sources for the studies you include. When studies are behind paywalls, use library access or request copies through document delivery. Share your full string and data extraction sheet in a public repository when the journal allows it.
Final checks before submission
Rerun the PubMed search on the day you submit and add any new records that meet criteria. Confirm that counts in text, tables, and the flow diagram agree. Check that every claim in the abstract appears in the main text and is backed by data. Run a spell check on all author names, trial IDs, and numbers.
Useful official resources
Keep handy tabs open while you work. The PubMed Help pages describe filters, fields, and the Advanced builder in plain language: PubMed Help. PRISMA 2020 hosts the checklist and flow diagrams that guide clear reporting. For protocol registration, see PROSPERO.