Protocol Amendments and Deviations in Systematic Reviews: What to Record and Report
An amendment is an approved change to the planned methods of a systematic review. A deviation is a point where the review was actually conducted differently from those methods. Whichever term applies, keep the original plan on record, say plainly what changed, date the decision, explain why it happened, name who signed off on it, note what the team already knew at that moment, weigh its effect on the review, and report it in the registry and the final paper wherever that’s expected.
A protocol is really just a public promise: here’s what we intend to do, written down before the evidence has any chance to nudge the plan. Real reviews rarely go exactly to script. Data go missing, an unfamiliar study design shows up, an assumption doesn’t hold, and reviewers have to adapt an eligibility rule, rework an analysis, or quietly drop a method they’d originally planned to use.
A change on its own isn’t proof of bias. What readers actually need is enough detail to judge its timing, its reasoning, and what it did to the results. The real risk shows up when a team swaps an outcome, a subgroup, an eligibility rule, or an analysis after already seeing how that swap would affect the findings. Without a written record, there’s simply no way to tell a reasonable adaptation apart from a result-driven one.
PRISMA-P asks protocol authors to say up front how they’ll document any important amendments.1 PRISMA 2020 then expects the finished review to describe and explain whatever changed from the registration or protocol.2 Cochrane’s instruction lands in the same place: declare departures from prespecified methods and explain them once the review is done.3
Amendment, deviation, clarification, or non-implementation?
Classify changes the same way throughout the project. Write the rule into the protocol or your project procedures, then stick to it from approval all the way through manuscript submission.
| Event | Working definition | Typical example | Required response |
|---|---|---|---|
| Amendment | The team approves a substantive revision to a planned method, ideally before actually using the revised version. | A database goes offline before the final search runs, so the team signs off on a replacement source and updates the search plan. | Keep the original plan, date and approve the revision, update the public record, then report it in the review. |
| Deviation | What actually happened during the review differs from the prespecified method, whether the team notices before or after the fact. | Duplicate extraction was planned with two reviewers, but one outcome ended up extracted by one person and just checked by another. | Note what happened, why, who accepted it, and whether it could touch the evidence or the conclusions. |
| Clarification | The team spells out an existing rule more clearly without changing what it actually means. | An age criterion already given in years gets rewritten to explain how mixed-age samples should be handled. | Keep the wording history. Check whether the clarification actually changes earlier decisions; recheck records if it does. |
| Correction | The team fixes a plain factual or clerical error in the protocol or registration. | A database name or an author’s affiliation was typed incorrectly. | Correct the public record where you can, and leave a dated note if the correction might otherwise confuse readers. |
| Non-implementation | A prespecified method just doesn’t get used, because the evidence doesn’t support it or its trigger condition never happened. | Meta-analysis was planned, but the included studies turn out too clinically diverse to pool. | State the planned method, the constraint that got in the way, and what was done instead. Don’t present the unused method as though it had been completed. |
A clarification can slide into amendment territory the moment it changes which studies, outcomes, or analyses actually qualify. Look at the effect it has, not just the label someone slapped on the edit.
Readers need to know whether the evidence shaped the method
Prespecification exists to limit how much freedom reviewers have to pick methods after they already know which choice gives them the answer they’d prefer. That protection quietly falls apart the moment reviewers can revise outcomes, eligibility rules, or analyses without leaving any dated trail.
Kirkham and colleagues compared 288 Cochrane protocols against their finished reviews. Sixty-four pairs, 22%, differed on at least one outcome measure. Only four of those 64 reviews actually explained the change in the review text.4 That study is dated and comes from one specific review environment, but it still captures the core problem plainly: readers can’t evaluate a change the report never even admits happened.
AMSTAR 2 treats prospective methods and justification of significant deviations as a critical domain when appraisers score systematic reviews of interventions.5 A protocol amendment record isn’t just a compliance checkbox, then. It’s the trail that shows how the final methods actually came to be.
Constraint known before results
The team learns before screening starts that a planned database is no longer accessible. It writes down the problem, agrees on a replacement source with an information specialist, and updates the registration before running the final searches.
Choice made with outcome knowledge
The team switches the primary time point after noticing that a different one produces a stronger effect. The report hides the original plan and presents the new choice as though it had been prespecified all along.
What to record for every amendment or deviation
Write the entry the moment the issue comes up. A note reconstructed weeks later while drafting the manuscript will miss exactly the detail that matters most: what the team actually knew at the time it made the call.
- Stable identifier. Give each event its own number so meeting notes, registry updates, and manuscript text can all point back to the same decision.
- Original method. Quote or cite the exact protocol text and keep its version intact.
- Revised or actual method. State plainly what changed or what actually happened in practice.
- Date and review stage. Note when the issue came up, when it was approved, and which tasks had already started.
- Trigger. Name the constraint, the new information, the error, or the event that set this off.
- Knowledge state. Say whether reviewers had already seen study identities, outcome data, effect estimates, or pooled results.
- Rationale. Explain why the revised approach was methodologically better or practically unavoidable.
- Decision authority. Name who proposed it, who reviewed it, and who approved it, including any independent input sought.
- Affected materials. List every protocol, registry entry, search record, form, analysis plan, and manuscript section that needs updating.
- Impact assessment. Say whether the event could shift eligibility, data, risk of bias, synthesis, certainty, or conclusions.
- Corrective action. Note any re-screening, re-extraction, re-analysis, or sensitivity check the change requires.
- Reporting location. Note exactly where readers will find this event in the registry, the review, and any supplement.
The paired Protocol amendment and deviation log turns these fields into a working document you actually fill in. This guide walks through the judgement behind each field; the resource is where the decisions live while the review is underway.
A practical workflow for reviewing a proposed change
Pause the affected task
Stop different reviewers from quietly applying different rules to the same problem. Save the current protocol, forms, and analysis files as they stand. Describe the issue without touching the original record.
Compare the plan with the observed problem
Find the exact protocol statement governing this task. Write down the practical evidence that exposed the problem, whether that’s a pilot decision, missing data, or an incompatible design.
Declare what the team has seen
Outcome knowledge changes how risky a decision is. State plainly whether the team has already seen study identities, directions of effect, statistical significance, or synthesis results. Bring in independent methodological review for anything high-risk.
Assess the methodological consequences
Trace the proposed change through eligibility, searching, selection, extraction, appraisal, synthesis, and reporting. Decide whether completed work now needs redoing. Even a narrow wording change can force a full recheck if it shifts how reviewers classify records.
Approve and version the decision
Use whatever authority the protocol or project governance plan actually specifies. Record any dissent and any conflicts of interest. Give the revised method a new version number and an effective date.
Update public and working records
Follow the registry’s current process exactly. PROSPERO separates minor amendments from major revisions, and most method changes fall into the major category.6 OSF keeps submitted registrations intact and uses a transparent update process that requires justification.7 Make sure local forms and instructions match the approved version.
Implement, check and report
Finish any corrective work, then verify every affected output actually uses the same method. Carry the event into the final review and its supplement. Keep the link between the public change record and the manuscript description intact.
Changes that deserve close methodological review
Every substantive event belongs in the record. The changes below can shift which evidence makes it into the review, or which result ends up in the spotlight.
| Area | Example | Main concern | Useful response |
|---|---|---|---|
| Question and scope | Population or intervention boundaries shift after screening has already started. | The change can add or drop evidence in a direction tied to the findings. | Record the evidence already known, reapply the final rule to every record, and show what it did to inclusion. |
| Outcomes and time points | A secondary outcome gets promoted to primary, or the team picks a different follow-up window. | Reviewers might be steering toward the outcome or time point that gives a preferred result. | Keep the original hierarchy, explain the change, and report the prespecified analysis wherever it’s still possible. |
| Information sources | A database, registry, date limit, or language restriction changes. | Relevant studies might become more or less likely to enter the review. | Document coverage, dates, and rationale. Update the full search record and assess the risk of missing evidence. |
| Selection process | Independent duplicate screening drops to single screening with a verification step. | Error control no longer matches what was approved. | Explain the constraint, define what verification means here, and audit a sample or the full set as needed. |
| Data and appraisal | Extraction fields or the critical appraisal tool change after the included studies are already known. | The new method may end up fitting the observed evidence a little too conveniently. | Apply the revised method consistently, keep earlier judgements on file, and arrange independent review. |
| Synthesis | The effect measure, model, subgroup, or sensitivity analysis shifts after results are already visible. | Analytical flexibility here can move magnitude, precision, or interpretation. | Label post hoc analyses clearly, give the rationale, and compare against the prespecified analysis where it’s still valid. |
| Certainty and conclusions | The certainty approach gets skipped, or the decision threshold changes. | Readers might end up with a stronger conclusion than the planned method actually supports. | Explain the departure and what it did to the language used in the conclusions. |
Changes made after the team already knows the results need a direct, honest account of that knowledge. Cochrane warns that decisions like excluding certain studies or switching statistical analysis after seeing their effect are especially vulnerable to bias.8 If a post hoc analysis genuinely answers a useful question, label it exploratory and show exactly how it relates to the prespecified one.
Four examples from review practice
A planned database becomes unavailable
The team hasn’t run its final searches yet. It records the loss of access, compares alternative coverage with an information specialist, approves a replacement, and updates the registry. That’s an amendment. The final review names both the original plan and the revised source set.
A pilot exposes an ambiguous mixed-population rule
Reviewers disagree over studies with partly eligible samples, and the protocol never defined a threshold. The team sets a rule before formal screening starts and rechecks the pilot set. Call this an amendment if the rule changes which studies qualify. It’s only a wording clarification if the existing criterion already drove the same decisions.
A subgroup appears after heterogeneity is observed
The protocol never prespecified this subgroup, and the team proposes it only after seeing study results. Record it as a deviation from the analysis plan. Explain the hypothesis, label the analysis exploratory, and don’t present the subgroup as confirmatory evidence.
A planned meta-analysis is not defensible
Included studies turn out to measure different constructs with incompatible designs. The team keeps the prespecified synthesis plan on record, explains why pooling would mislead readers, and runs a structured synthesis without meta-analysis instead. Report both the non-implementation and the replacement method.
Where to report protocol changes
One sentence tucked into the Discussion rarely gives readers enough to work with. Put each part of the record where it actually does something useful.
| Location | What readers should find |
|---|---|
| Protocol history | The original method, the amendment date, the revised text, the rationale, the approver, and the current protocol version. |
| Registry or registration update | A public revision trail matching the approved change, pointing to a detailed protocol or amendment record when one exists. |
| Methods section | The methods actually used in the completed review, written as conduct rather than intention, with important protocol departures clearly flagged. |
| Protocol and registration subsection | The protocol citation, registration identifier, access link, and a concise statement of amendments. |
| Results or flow information | Any shift in record counts, included studies, available outcomes, or analyses that the event caused. |
| Discussion | What a consequential deviation did to limitations, certainty, interpretation, and applicability. |
| Supplement | The complete change log when the main report simply can’t fit the detail. |
PRISMA 2020 item 24c asks authors to describe the amendment, explain the reason, and say at what stage it happened.9 Cochrane asks authors to describe and explain every amendment to prespecified methods in the completed review.3 Journal headings vary, so follow your target journal’s instructions while still keeping the full decision history intact somewhere.
Common reporting mistakes
Overwriting the protocol
Editing the original file with no version history erases the comparison between planned and revised methods. Keep the approved version intact and create a dated successor instead.
Recording only the new method
Readers need the original plan, the changed method, and the reason for the switch. The final method alone doesn’t show the decision behind it.
Using “feasibility” as the whole rationale
Name the actual constraint. Say whether the problem was time, staffing, data availability, software, access, or a bad assumption. Explain why the chosen response actually protects the review.
Hiding analyses that were not performed
Report a planned subgroup, sensitivity analysis, or meta-analysis that couldn’t be completed. Give the exact condition that failed and what was used instead.
Calling a substantive change a clarification
Check whether the edit actually changes decisions. If a study, outcome, or analysis would now be treated differently, it’s a methodological revision, not a clarification.
Updating one record only
A registry update, protocol, working form, and manuscript can drift apart fast. Assign one owner to reconcile every affected document before the next review stage begins.
Audit the protocol against the completed review
Compare the final review against the last approved protocol and registration, line by line rather than from memory. Check the question, eligibility criteria, outcomes, information sources, selection process, extraction, appraisal, synthesis, and certainty methods. PRISMA-P gives you a consistent protocol baseline for this comparison.10 Empirical work on selective inclusion and reporting explains why this comparison needs to cover outcomes and analyses, not just wording.11
For each difference you find, track down a dated entry and confirm the registry and the review report actually agree. Add any event that’s missing from the record. If an entry is reconstructed after the fact, say so plainly. Don’t assign it an earlier approval date than it deserves.
Finish the audit by checking effect, not just paperwork. Revisit decisions made under an older rule. Confirm sensitivity analyses are running on the correct datasets. Read the abstract and conclusions against the final methods, so a late change doesn’t leave the detailed record out of sync with the public summary.
Keep one record from the first change to final reporting
The Protocol amendment and deviation log holds the original plan, the revised or actual method, dates, rationale, approval, knowledge of results, consequences, and reporting location. Use it while the review is underway, then carry the verified entries into the registry, manuscript, and supplement.
Document protocol changes →Related protocol planning resources
Frequently asked questions
Can I change a systematic review protocol after registration?
Yes. A necessary change is fine after registration as long as the team documents and justifies it properly. Keep the original version, record the date and review stage, update the registry through its current process, and report the amendment in the completed review. State exactly what evidence the team had already seen when it approved the change.
What is the difference between a protocol amendment and a protocol deviation?
An amendment is an approved revision to the planned method, ideally made before the revised method is actually used. A deviation describes conduct that ended up differing from the approved plan. Different organizations define these terms differently, so include the original method, the actual or revised method, timing, rationale, approval, and effect regardless of which label you use.
Do I report a planned analysis that could not be performed?
Yes. Name the planned analysis, explain why the available evidence couldn’t support it, and describe what was used instead. A team might, for example, drop a planned subgroup analysis because too few studies provide suitable data. The final report shouldn’t imply the analysis was completed when it wasn’t.
Does every protocol correction need a manuscript statement?
A clerical correction with no real effect on the method might only need a transparent version note or a registry correction. Report it in the manuscript if it could affect interpretation, reproducibility, eligibility, or analysis. Keep an internal record either way, so the team can justify treating it as a mere correction.
Where should protocol amendments appear in a systematic review?
Update the registration or protocol history wherever the platform allows it. In the completed review, cite the protocol, describe important amendments in the Methods or a designated differences section, report their consequences in the Results, and discuss any effect on limitations or certainty. Use a supplement for the full log when it’s needed.
Does a protocol deviation invalidate a systematic review?
No single deviation decides that on its own. Look at the timing, the rationale, whether results were already known, how important the change is, and what corrective action followed. A transparent response to something unforeseen can actually protect the review. An unexplained, result-driven change to outcomes or analyses is where the real concern lies.
Methodological guidance and empirical evidence
- Shamseer L, Moher D, Clarke M, Ghersi D, Liberati A, Petticrew M, et al. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015: elaboration and explanation. BMJ. 2015;350:g7647. doi:10.1136/bmj.g7647
- Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. doi:10.1136/bmj.n71
- McKenzie JE, Brennan SE, Ryan RE, Thomson HJ, Johnston RV, Thomas J. Chapter III: Reporting the review. In: Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA, editors. Cochrane Handbook for Systematic Reviews of Interventions. Version 6.5. Cochrane; 2024. Current online chapter
- Kirkham JJ, Altman DG, Williamson PR. Bias due to changes in specified outcomes during the systematic review process. PLoS One. 2010;5(3):e9810. doi:10.1371/journal.pone.0009810
- Shea BJ, Reeves BC, Wells G, Thuku M, Hamel C, Moran J, et al. AMSTAR 2: a critical appraisal tool for systematic reviews that include randomised or non-randomised studies of healthcare interventions, or both. BMJ. 2017;358:j4008. doi:10.1136/bmj.j4008
- Centre for Reviews and Dissemination. PROSPERO frequently asked questions. University of York. Current guidance
- Center for Open Science. Welcome to Registrations and Preregistrations. OSF Support. Current guidance
- Higgins JPT, Lasserson T, Chandler J, Tovey D, Thomas J, Flemyng E, Churchill R. Chapter 1: Starting a review. In: Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA, editors. Cochrane Handbook for Systematic Reviews of Interventions. Version 6.5. Cochrane; 2024. Current online chapter
- Page MJ, Moher D, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. PRISMA 2020 explanation and elaboration: updated guidance and exemplars for reporting systematic reviews. BMJ. 2021;372:n160. doi:10.1136/bmj.n160
- Moher D, Shamseer L, Clarke M, Ghersi D, Liberati A, Petticrew M, et al. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015 statement. Syst Rev. 2015;4:1. doi:10.1186/2046-4053-4-1
- Page MJ, McKenzie JE, Kirkham J, Dwan K, Kramer S, Green S, Forbes A. Bias due to selective inclusion and reporting of outcomes and analyses in systematic reviews of randomised trials of healthcare interventions. Cochrane Database Syst Rev. 2014;(10):MR000035. doi:10.1002/14651858.MR000035.pub2
Scope note: This guide supports systematic review documentation and reporting. Registry functions and journal requirements change, so check the current instructions for the platform and publication venue used by your review.
