12 checks before you hit submit.

The 12-point checklist

Work through it in order. Anything you cannot answer cleanly is worth fixing before you submit rather than explaining to a reviewer afterwards.

1. 1 The question, the analysis and the conclusion still match Read the abstract's conclusion, then the primary analysis, then the research question. Papers drift during writing. If the conclusion answers a slightly different question from the one the analysis addresses, a reviewer will notice before you do.
2. 2 The right reporting guideline is completed, not just cited CONSORT for trials, STROBE for observational studies, PRISMA for reviews, STARD for diagnostic accuracy, TRIPOD+AI for prediction models, CARE for case reports. Complete the checklist with page numbers. Many journals now screen against it automatically.
3. 3 Ethics approval, consent and registration are stated specifically Name the approving committee and the approval reference. State how consent was obtained or why it was waived. Give the registration number and registry for trials and, increasingly, for systematic reviews. Vague ethics statements are a common desk-rejection trigger.
4. 4 Sample size is justified — prospectively or honestly For a planned study, give the calculation and its assumptions with their source. For a retrospective study, say what precision the available sample provides rather than performing post-hoc power analysis on the observed effect, which is uninformative.
5. 5 The methods section describes what was actually run Every analysis in the results should appear in the methods, and every method described should produce something in the results. Name the software and version. State the significance level and whether tests were one- or two-sided.
6. 6 Effect sizes and confidence intervals appear everywhere a p-value does A p-value alone tells the reader nothing about magnitude or precision. Report mean differences, odds ratios, risk ratios or hazard ratios with their confidence intervals — including for the results that were not significant.
7. 7 Missing data are quantified, explained and handled State how much was missing per key variable, what pattern it followed, and what you did about it. "Cases with missing data were excluded" without numbers is not a missing-data strategy, and reviewers increasingly reject it as one.
8. 8 Model assumptions were checked and the check is reported Normality, homoscedasticity, independence, linearity, multicollinearity, proportional hazards, proportional odds — whichever apply. Say that you checked, say what you found, and say what you did when an assumption failed.
9. 9 Confirmatory and exploratory analyses are clearly separated Label pre-specified analyses as pre-specified and post-hoc analyses as post-hoc. State how multiplicity was handled across multiple outcomes, subgroups or comparisons. An exploratory subgroup finding presented as confirmatory is the fastest route to a hostile review.
10. 10 Tables and figures stand alone, and their numbers match the text Each should be interpretable without the body text: defined abbreviations, units, denominators, and a stated statistical test. Then check every number in the abstract against the table it came from. Mismatches are common and they damage credibility disproportionately.
11. 11 The discussion claims only what the design supports Observational data support association, not causation, and the wording should reflect that. State the limitations that actually matter — unmeasured confounding, selection, generalisability — rather than a generic paragraph. Reviewers trust a paper that names its own weaknesses.
12. 12 Authorship, conflicts, funding and data availability are complete ICMJE-compliant authorship with stated contributions, declared conflicts for every author, funding source and role, and a data-availability statement. These are administrative, they take ten minutes, and their absence gets papers returned before anyone reads the science.

The three that end most papers early

If you only have time for three, make them these — they are what an editor screens for in the first few minutes:

• Statistical reporting gaps. Effect sizes without confidence intervals; missing-data handling undescribed; methods that do not match the results.
• Incomplete ethics documentation. No committee named, no approval reference, no consent statement.
• Unclear authorship. Contributions, conflicts and funding left vague or absent.

If it still comes back

Rejection is a diagnosis problem, not a dead end. Assess the manuscript across four separate dimensions — scientific, methodological, statistical, presentational — and separately ask whether the journals you have been submitting to were ever appropriate for its scope and methodology. One manuscript that reached us after more than ten rejections was accepted by the first journal the revised version was sent to. That case is documented here.