“The statistical analysis should be reviewed by a qualified statistician.” The decision letter says it; you do not have one. What the comment usually means, the real options, and how to answer it honestly.
The sentence reads like a verdict, but it arrives for three different reasons, and the right response depends on which one you have. Sometimes it is specific: the reviewer has named concrete concerns — the model, the missing-data handling, the multiplicity — and wants an expert to adjudicate them; the list is usually right there in the adjacent comments. Sometimes it is policy: some journals route certain designs to statistical review as a matter of course, and the request is procedural rather than an indictment of your analysis. And sometimes it is a hedge: a clinical reviewer signaling that the statistics are outside their own comfort zone, deferring rather than critiquing. Read the whole letter before deciding what you are actually being asked for — a named expert, specific fixes, or both.
1. Bring in a biostatistician — often as a co-author. If the comments call for reanalysis, a different model, or judgment calls you cannot defend alone, this is the answer, and it is worth doing properly: a statistician who reanalyzes data, shapes the interpretation, and drafts the methods revision has usually earned authorship under your journal's criteria, and a named statistical co-author is the strongest possible response to the comment. The constraint is access — departmental statistical support is scarce and queued, which is exactly why the comment is so common. The categories and trade-offs are compared in your options for statistical review before submission.
2. Commission an independent statistical review. Paid manuscript-level statistical review by a contracted expert typically runs $150–$400 at published list prices, returned over days. For a high-stakes revision with no local access, buying human judgment can be the right call — and it gives you a concrete, honest sentence for the response letter.
3. Fix the named problems, point by point. When the request is a hedge attached to a short list of specific concerns, the productive reading is that the concerns are the assignment. Added sensitivity analyses, corrected reporting, and recalibrated conclusions — each answered in the letter with page and line — often resolve the comment without any new person, because what the editor needs is evidence the statistics were taken seriously. The playbook for those answers is responding to Reviewer 2's statistics comments.
4. Use an automated review to prepare — knowing exactly what it is. A structured pre-submission statistical review recomputes what the reported numbers allow, reads design–claim fit and conclusion calibration with two independent engines, and shows its work — so the revision that goes back has the checkable problems already found, and if you do engage a statistician, their hours go to judgment instead of arithmetic. Stated plainly: it is not a statistician, and it does not satisfy a journal's requirement for one. If the editor requires named statistical review, an automated report cannot be represented as that — it flags; it never certifies.
Whichever route you take, the rule is the same: describe exactly what happened, and never imply expert review that did not occur — a response letter is part of the scholarly record, and misstating who reviewed what is the kind of claim that unravels badly. Honest model language for each case:
“A biostatistician (Dr. X, now a co-author) has re-run the primary analysis using a mixed-effects model, revised the Methods (pp. 5–7), and reviewed the full manuscript.” — when you added the expert.
“The statistical analysis was independently reviewed by a commissioned biostatistical service; in response we corrected the denominator in Table 2 and now report exact p-values throughout.” — when you commissioned the review.
“We have addressed each statistical concern individually: the multiplicity comment with an explicit exploratory labeling (p. 6), the missing-data comment with a sensitivity analysis (Table S3), and the null-result wording throughout (p. 9).” — when you fixed the named problems. If an automated review helped you find or check those fixes, the letter describes the fixes; the tool is preparation, not a credential to cite.
However the statistician question resolves, the revised draft should not go back unchecked — revisions introduce new inconsistencies while fixing old ones. RigorMD re-reviews a revised manuscript for $15 (when it was first reviewed here; $30 for a manuscript new to it) with a finding-by-finding comparison against the first report, and if you paste the decision letter, the report classifies every reviewer point must-fix versus presentational ask with a draft response to edit. How that works is on answer the reviewers with a checked revision →. It flags for your judgment; it does not certify, write your letter, or promise the outcome — and on any complex revision, a statistician or mentor should still read the final response.