PRISMA 2020 in Practice: Five Things Most Systematic Reviews Still Get Wrong

PRISMA 2020 has been the reference reporting standard for systematic reviews for several years. Most journals require it; most reviewers check the checklist. Yet a striking proportion of published reviews still fall short in ways that are easy to identify and easier to avoid. After running many reviews across therapeutic areas, here are the five most common failures and how to design around them.

1. The PROSPERO Step is Skipped or Backdated

Item 24a asks where the protocol is registered; the honest answer is too often "nowhere" or "registered after extraction began." Registration is a pre-specification tool — register before screening starts, document the date, and register an amendment if the protocol changes.

2. The Search Strategy is Not Reproducible

"We searched PubMed and Embase using relevant terms" is a description, not a strategy. Item 7 requires the full search string for every database, the run date, and any filters, so a reader can re-run and reproduce your result.

3. Dual Screening is Described But Not Actually Performed

"Two reviewers independently screened" is one of the most over-used sentences in evidence synthesis, and the audit trail is rarely present. Use Rayyan, Covidence, or DistillerSR with blinded screening and conflict logs, and report inter-rater agreement.

4. Risk-of-Bias Assessment is Treated as a Checkbox

Item 11 requires interpretation of risk of bias in the synthesis, not just a table. Downgrade certainty where risk is high and run sensitivity analyses excluding high-risk studies.

5. GRADE Certainty is Omitted or Applied Superficially

For any review meant to inform clinical or policy decisions, GRADE is essential — per outcome, with documented engagement of each domain (risk of bias, inconsistency, indirectness, imprecision, publication bias). That is what makes a review usable downstream.