PICO Question: Build a PubMed Search with MeSH Terms

A PICO question splits a clinical problem into four parts, and each part becomes one block of a PubMed search. This guide takes one atrial fibrillation question to a search string you can paste.

What is a PICO question?

A PICO question names the Patient or problem, Intervention, Comparison, and Outcome. Richardson and colleagues described "The well-built clinical question" in ACP Journal Club in 1995, and a study of PICO-based PubMed searching cites it.

Two variants add one element each. A UNC guide describes PICOT as adding time, and a Duquesne guide describes a study design element (PICOS). Plain PICO covers most quick searches.

How do you write a PICO question?

Write one line per element, then join them into a single question. Keep the outcome clinical (stroke, bleeding, death) and skip surrogate markers when you can.

Worked example, starting from the question "DOAC or warfarin for AF?":

• P: adults with atrial fibrillation
• I: direct oral anticoagulants (apixaban, rivaroxaban, dabigatran, edoxaban)
• C: warfarin
• O: stroke or systemic embolism, and major bleeding

Final question: in adults with atrial fibrillation, do DOACs compared with warfarin change stroke or systemic embolism and major bleeding? The 2023 ACC/AHA/ACCP/HRS atrial fibrillation guideline, as summarized by the ACC, says direct oral anticoagulants are preferred over warfarin except in patients with mitral stenosis or mechanical heart valves. Search when you want the trial data behind that statement. For stroke risk, use the CHA2DS2-VASc calculator.

How do you turn PICO into a PubMed search?

Build one OR block per PICO element, then join the blocks with AND. Use a MeSH heading plus [tiab] keywords in each block, because each catches records the other misses.

1. Pick MeSH headings. NLM describes MeSH as a controlled, hierarchically organized vocabulary. A MeSH term automatically includes the more specific terms beneath it, and [mh:noexp] turns that off. Borrow headings from a known good paper: the 2014 Lancet meta-analysis of the four pivotal DOAC trials lists Atrial Fibrillation, Warfarin and Anticoagulants.
2. Add [tiab] synonyms. Filters based on indexing may miss citations that haven't finished MEDLINE indexing, so free text covers new papers.
3. Join with uppercase Boolean operators. Use OR within a block, AND between blocks, and parentheses to group concepts.
4. Leave the outcome out. Abstracts phrase outcomes many ways, and adding them early can drop good papers.
5. Filter instead. Pick Meta-Analysis or Randomized Controlled Trial in the sidebar (most use the publication type tag [pt]). Or run the string through Clinical Queries, Therapy category, Broad scope, which "returns more results" and is less specific.

("Atrial Fibrillation"[mh] OR "atrial fibrillation"[tiab]) AND (apixaban[tiab] OR rivaroxaban[tiab] OR dabigatran[tiab] OR edoxaban[tiab] OR "direct oral anticoagulant"[tiab] OR "direct oral anticoagulants"[tiab] OR DOAC[tiab] OR DOACs[tiab]) AND (warfarin[mh] OR warfarin[tiab] OR "vitamin K antagonist"[tiab] OR "vitamin K antagonists"[tiab])

Check Search Details in History on the Advanced page to confirm PubMed translated your tags as expected. Field tags and phrase searches bypass automatic term mapping, so tagged terms mean exactly what you typed.

How do you read the results fast?

Check publication type, then population, then result and harm, then risk of bias. PubMed's default Best Match sort re-ranks results with a machine learning model, so the first page deserves your attention.

1. Publication type. The Lancet record carries the tags Meta-Analysis and Comparative Study, which tells you the design before you read a word.
2. Population. That meta-analysis pooled 71,683 patients from four trials (RE-LY, ROCKET AF, ARISTOTLE, ENGAGE AF-TIMI 48). Compare that population with yours.
3. Result and harm. It reported about 19% less stroke or systemic embolism with DOACs, more gastrointestinal bleeding, and lower all-cause mortality.
4. Risk of bias. For trials, glance at the five RoB 2 domains: randomization, deviations from intended interventions, missing outcome data, outcome measurement, and selection of the reported result.

Where does this fall short?

A first-pass PubMed search finds papers and does not rank their quality. Indexing lag can hide new papers from MeSH searches, and the example meta-analysis dates from 2014, so newer trials and subgroups (older adults, kidney disease, obesity) need their own searches. This article checked the ACC summary of the 2023 guideline on 2026-10-10, and a newer focused update may exist. For anything that changes management, read the full guideline and the primary trials.