1. Searching a single database
One database means one indexing vocabulary and one set of included journals. Anything outside it is invisible, and the omission is obvious to anyone who checks. Search at least two bibliographic sources plus a trial registry, and record each search string, the source, the date, and the number of hits.
2. Skipping grey literature
Theses, conference abstracts, regulator documents and unpublished trial results carry a different bias profile from journal articles — typically less favourable results. Leave them out and your synthesis leans positive. Search registries and regulator archives, and state what grey literature you looked for even where you found nothing.
3. Not registering a protocol
Without a dated protocol there is no evidence your criteria preceded your results. That leaves every later decision open to the charge of being fitted to the findings. PROSPERO registration is free and fast; do it before screening starts.
4. Selective outcome reporting
Deciding which outcomes matter after seeing which ones were significant is one of the clearest integrity failures in evidence synthesis. Fix your primary outcome in the protocol. If you add an outcome later, label it post hoc and explain why.
5. Treating a registry entry as a publication
A ClinicalTrials.gov or ISRCTN record describes an intended or completed study; it is not a peer-reviewed report of results. Registry entries are valuable for finding unpublished work and detecting outcome switching, but never extract them as if they were papers. Where a registry record is your only source, mark it as such.
6. Ignoring retractions and corrections
Retracted studies continue to be cited for years. Check every included record against retraction notices and published corrections before extraction, and re-check shortly before submission. A retracted study inside a meta-analysis undermines the whole result.
7. Excluding studies without recording why
PRISMA 2020 expects a reason for each full-text exclusion. Reconstructing those reasons from memory weeks later is unreliable and shows. Record the reason at the moment you decide — wrong population, wrong comparator, no extractable outcome, duplicate report, full text unavailable.
A short pre-submission check
- Is every search string, source and date recorded and reproducible?
- Did you search beyond journal articles?
- Is the protocol registered and dated before screening?
- Do the reported outcomes match the protocol?
- Is every registry-only record labelled as such?
- Have all included studies been checked for retraction?
- Does every full-text exclusion carry a reason, and do the numbers add up in your PRISMA flow?