Amending a systematic review protocol means making a documented, dated change to your registered methods after the protocol is live, and recording why. Amendments are not a sign of poor planning; they are expected as real evidence is encountered, and the only rule that matters is that every change is transparent and traceable rather than hidden.
Why an honest amendment beats a perfect-looking record
Reviewers and editors do not expect a protocol to survive contact with the evidence untouched. What erodes trust is a final review whose methods quietly differ from its registration with no explanation, because that reads as outcome switching. A clear amendment log does the opposite: it shows you adapted for a methodological reason and stood behind the change. This is the same principle that justifies registering the protocol in the first place, and it is why platforms keep every prior version visible, as we cover in PROSPERO versus OSF registration.
Changes that usually require an amendment
Eligibility and scope
If you tighten or loosen your inclusion and exclusion criteria, add a study design, or shift the population, that is a substantive change and needs recording. These often surface during piloting, when real abstracts expose ambiguity in criteria that looked clear on paper.
Search and outcomes
Adding a database, revising the query after a PRESS peer review of the search, or changing which outcome is primary all belong in the amendment log. Outcome changes are the most scrutinised of all, so document the reason with particular care.
Analysis and synthesis
Switching from a planned meta-analysis to a narrative synthesis because the studies proved too heterogeneous, or adding a subgroup analysis you did not pre-specify, are both amendments. A post-hoc subgroup analysis is allowed, but it must be flagged as post-hoc, never dressed up as planned. Changing the statistical model, for instance moving from a fixed-effect to a random-effects model once heterogeneity proves substantial, sits in the same category: defensible on a stated methodological reason, indefensible if introduced silently to shift the result.
How to record an amendment properly
Good amendment records share four features. Each entry has a date, a description of what changed, the reason for it, and a note of when in the process it happened, especially whether it was before or after you had seen any results. The timing is what separates a legitimate refinement from a results-driven one. Update the registration entry itself, keep a running amendment table in your protocol document, and carry that same table into the methods section of the final manuscript so the published review and the registration tell one story.
The discipline that makes this painless is to log the change at the moment it is decided, not at write-up. A team that reconstructs its amendments months later, from memory, almost always blurs the timing, and timing is the one detail a reviewer can check against the dated registration history. Keep the amendment table beside the protocol from day one, and treat a new row as part of making the decision rather than a chore that follows it. The same habit pays off when the review is reported against the PRISMA 2020 checklist, which asks explicitly what was changed and why, so a maintained log becomes a paste-ready answer rather than a forensic exercise.
A worked amendment-log entry
Abstract advice is easy to nod along to and hard to apply, so here is what a defensible entry actually looks like in practice. Imagine that during piloting the extraction form you discover that several eligible trials report a continuous outcome your protocol assumed would be binary. A clean log entry reads:
- Date: the calendar date the decision was made, not the date you wrote it up.
- Change: “Added standardised mean difference as an effect measure alongside the planned risk ratio, to accommodate continuous outcome reporting.”
- Reason: “Three included trials reported the outcome continuously; pooling them required a continuous effect measure.”
- Timing: “Identified during pilot extraction, before any outcome data were synthesised.”
That last line is what does the heavy lifting, because it places the change before the synthesis saw any result. Knowing the difference between an effect measure choice made for a methodological reason and one made to flatter a finding is exactly what a reviewer is testing when they compare your registration to your manuscript.
Changes that do not need an amendment
Not every edit rises to the level of an amendment, and over-logging trivia buries the changes that matter. Correcting a typo, fixing a broken reference, clarifying wording without altering meaning, or updating a database’s interface name are housekeeping, not amendments. The test is whether the change could plausibly alter which studies are included or how the results are analysed. If it could, log it; if it could not, leave the audit trail uncluttered. Adding a post-hoc subgroup analysis always clears that bar, whereas reformatting a table never does.
Before versus after seeing results
The single most important distinction is timing. A change made before any data are extracted is a routine refinement and carries little weight against you. A change made after results are in is far more sensitive, because it can look like it was chosen to flatter the findings. That does not make late changes forbidden, but it raises the bar for justification, and it is one reason a generous, well-pilot-tested protocol up front saves trouble later, as we note in framing a strong research question. Transparency, not rigidity, is what keeps a review credible. When the review is finally written up against the PRISMA 2020 guideline, the same amendment table belongs in the methods section, so the published paper, the registration, and the protocol all tell one consistent story.