Responding to reviewers on a meta-analysis

Reading the decision

Start by reading the editor's letter, not the reviewers' comments. The editor decides, and the letter says what the journal expects: which issues are essential, which are optional, and how the revision will be judged. A "major revision" means that the paper may be acceptable if the main concerns are met. A "minor revision" means that only small changes are needed. "Reject and resubmit" signals that the paper would need substantial change and be considered again as a new submission. Check the deadline and the format required for the response.

Then read each reviewer's report twice. The first time, read for the overall impression and let any irritation pass. The second time, mark each comment, and sort them into groups: errors you must fix, requests for additional analyses, requests for clarification, requests that would change the scope, suggestions you can adopt and comments based on a misunderstanding. A misunderstanding usually means that the paper was not clear, so the fix is often a change to the text, not a rebuttal.

Structuring the response letter

A point-by-point response is the standard. It should be easy for the editor and the reviewers to follow, and complete: every comment answered, none skipped.

• Open with a short letter to the editor. Thank the editor and reviewers briefly, summarize the main changes in a few lines and note any new analyses.
• Quote each comment in full, in a distinct style such as italics or a colored font, and number them (Reviewer 1, comment 3).
• Answer directly. Begin with what you did: "We have added..." or "We respectfully disagree because...".
• State where the change is. Give the page, line or section, and quote the revised text where it is short.
• Keep a neutral, courteous tone. No sarcasm and no assumptions about the reviewer's motives.
• Supply two versions of the manuscript, one with tracked changes and one clean, as most journals request.

Keep answers proportionate. A trivial comment needs a short reply, and a substantive comment needs enough explanation for the reviewer to see that it has been taken seriously.

An illustration

The following is an illustrative exchange, written for this guide and not taken from a real review.

Reviewer 2, comment 4: The authors use a random-effects model but do not report a prediction interval. With only seven studies and substantial heterogeneity, the confidence interval may give a misleading impression of consistency.

Response: We thank the reviewer for this point and agree. We have added a 95 percent prediction interval to the Methods (page 8, lines 14 to 17) and to the Results (page 11, lines 3 to 6), calculated with a t distribution with k minus 2 degrees of freedom. The prediction interval is wider than the confidence interval and includes the null, and we have revised the Discussion (page 15, lines 22 to 28) to say that the effect in a new setting may be smaller or absent. We have also updated the abstract and Figure 3.

The response quotes the comment, concedes the point, says what was done and where, and states the consequence for the interpretation. It does not argue, and it does not hide the fact that the new analysis weakens the conclusion.

Common requests on meta-analyses

Some requests recur, and it is worth preparing for them. Many can be anticipated and addressed before submission.

• Heterogeneity. Report tau-squared and I-squared with intervals, add a prediction interval, and explore sources by subgroup analysis or meta-regression, within the limits of the number of studies. Explain why the model was chosen.
• Publication bias. Add a funnel plot, an asymmetry test suited to the effect measure and, if suitable, a sensitivity analysis such as trim-and-fill. With fewer than about ten studies, explain why such tests are not informative.
• Risk of bias. Use a recognized tool, present results by study and domain, and add a sensitivity analysis restricted to studies at lower risk of bias.
• Certainty of evidence. Add a GRADE assessment and a summary of findings table.
• Search. Provide the full strategy, add databases or registries, or update the search. If the search is out of date, rerun it and screen the new records.
• Model choice and estimator. State and justify the estimator and the confidence interval method, and perhaps show the fixed-effect result for comparison.
• Outlying or influential studies. Present leave-one-out analyses.
• Reporting. Complete the PRISMA checklist and flow diagram, and register or explain the lack of registration.
• Overstated conclusions. Tone down causal language and tie conclusions to the certainty of evidence.
• Language and presentation. Edit the language and reduce redundancy.

Doing new analyses responsibly

Reviewers often ask for analyses that were not in the protocol. It is reasonable to run them, and it is important to label them as post hoc in the paper, to say that they were requested in peer review and to update the log of deviations from the protocol. Report the results whatever they show. If a requested analysis weakens or reverses a finding, say so in the main text and adjust the conclusions. Selective reporting of only the analyses that support the preferred message is the same fault the review is meant to detect in the studies it includes.

Check the numbers carefully before submitting. New analyses mean new tables and figures, and a changed pooled estimate may be quoted in the abstract, the text, the summary of findings table and the conclusions. Update all of them, and reread the paper for consistency. A frequent embarrassment is a revised result in the Results section and the old one still in the abstract.

Disagreeing with a reviewer

Some comments will be wrong, or based on a different view of the method, or will ask for something outside the scope of the paper. Disagreement is allowed, if it is handled with respect and evidence.

• Start with what you agree with. If part of the comment is valid, say so.
• Explain the reason. Cite the Cochrane Handbook, a methods paper or the protocol, where it supports your choice.
• Show you considered it. If feasible, run the requested analysis and report it in a supplement, saying why it is not in the main text.
• Acknowledge it in the paper. If a reviewer raised a point that a reader might also raise, add a sentence to the limitations or methods.
• Do not ignore it. An unanswered comment will come back.
• Let the editor arbitrate when needed. If two reviewers disagree, state which approach you adopted and why.

Avoid language that attacks the reviewer. It never helps, and the editor reads the letter.

Managing the process

Organize the work as a short project. List all comments in a table with columns for the reviewer, the comment, the type, the action, who is responsible and the status. Estimate the time for new analyses, especially searches and data extraction, which can take weeks. If more time is needed, ask the editor for an extension before the deadline, with a reason, which is usually granted. Share the response with all co-authors, and obtain their approval before resubmission. After submission, if the journal asks for a second round, handle it in the same way, concentrating on the points still open.

Common mistakes

• Answering emotionally, in the first days after the decision.
• Skipping comments, or answering only the ones that are easy.
• Saying a change was made without saying where.
• Making a change in the text and not in the abstract, tables or figures.
• Adding analyses without labeling them as post hoc.
• Reporting only the new analyses that support the conclusions.
• Arguing with the reviewer instead of improving the clarity of the paper.
• Missing the deadline without telling the editor.

A second illustration: a disagreement

Again, the exchange is illustrative and not from a real review.

Reviewer 1, comment 2: The authors should have excluded all studies at high risk of bias from the primary analysis.

Response: We thank the reviewer. We agree that studies at high risk of bias can inflate effects and we examined this carefully. As stated in our registered protocol (section 7), the primary analysis includes all eligible studies, because restricting in advance can discard information when the number of studies is small, and we prespecified a sensitivity analysis restricted to studies at low or some concerns risk of bias. This analysis, in the Results (page 12, lines 9 to 15) and Supplementary Table 4, gives a similar estimate with a wider interval. We have added a sentence to the Discussion (page 16, lines 4 to 7) to make clear that the conclusion does not depend on the high-risk studies, and we have lowered the certainty of the evidence for risk of bias in the summary of findings table.

This response disagrees with the request in a respectful way, points to the protocol, shows that the underlying concern was addressed through a prespecified analysis and describes the resulting change in the paper.

The letter to the editor

The cover note to the editor is often read before anything else. Keep it to one page. Begin by thanking the editor and reviewers. Give a short list of the main changes in order of importance, for example the new prediction interval, the added GRADE assessment, the updated search and the revised conclusions. State clearly if the conclusions changed. Mention any comment on which you disagreed, so that the editor is not surprised, and refer to the detailed response. Confirm that all authors have approved the revision. If new authors or new contributors were added, explain the reason and obtain the journal's required agreement. The editor wants to see that the revision is complete, honest and easy to assess.

Integrity during revision

Revision is not a license to alter data or conclusions to please reviewers. All changes to extracted data, search results or analyses must be documented, and any correction of an error found during revision should be reported to the editor, with the effect on results. If an error is serious, an honest explanation is far better than a quiet fix. If AI tools are used to draft responses or edit text, follow the journal's policy on disclosure, and check every statement of fact and every number yourself, since a plausible but wrong figure in a response letter is easy to produce and hard to retract. Respect the confidentiality of the review process, and do not contact reviewers directly.

How we can help

We can work with authors to analyze the comments, run additional analyses such as prediction intervals, publication bias checks, GRADE tables and sensitivity analyses, revise the manuscript and prepare a point-by-point response with tracked and clean versions. [OWNER VERIFICATION REQUIRED] The relevant services are peer review revision support and statistical analysis.

Frequently asked questions