How should a medical manuscript be checked before submission?

A concrete task with RigorMed

1. Turn a pre-submission read-through into an itemised revision list

   These are suggested task inputs and outputs to request, not a new execution, a customer outcome or a guarantee of completion.

   Materials to provide: An existing abstract or manuscript, actual table and figure outputs, the applicable reporting checklist and the journal's current requirements.

   Requested outputs / delivery goal: A manuscript checking table with source locations, revision suggestions and questions for the authors to confirm.

2. An example request

   First compare sample sizes, units, key numbers and citation locations in the abstract and text. Create a checking table with original location, issue, evidence and suggested revision. Keep items without analysis outputs pending; do not invent results.

3. What still needs review

   This is an assisted pre-check, not a full peer review or permission to submit. Authorship, ethics, analysis facts and journal requirements still need individual confirmation.

Build a result-to-table/figure-to-text mapping first

Record sources separately for main and key secondary results. If original analysis outputs cannot be found, do not reconstruct or invent results from manuscript numbers.

For example, text reporting an adjusted risk ratio while a table lists an unadjusted odds ratio is an estimand mismatch, not a layout issue. Return to the frozen plan and outputs to determine what should be reported, then update every occurrence—not just one.

Check four evidence chains item by item

1. Verify every key number against analysis outputs

   Compare sample sizes, denominators, event counts, effect estimates, 95% confidence intervals or other uncertainty measures, units and time points. Ensure exploratory or sensitivity analyses are not presented as prespecified primary results and a nonsignificant result is not described as proof that groups are identical.

2. Trace each result through abstract, text, tables and figures

   List every occurrence of a main result and align populations, subgroups, variable names, measurement units and decimal precision. Tables need self-contained titles and notes. Figure axes, legends, denominators and error bars must correspond to actual data. Label schematic illustrations as non-data figures; they must not stand in for measurements or statistical results.

3. Open the original source for every key citation

   Verify authors, title, year and version using PMID, DOI, journal pages or official documents, then read the passage relevant to the claim. Database search results, abstracts and AI-generated citations cannot replace original-source verification. When only an abstract is accessible, narrow or remove claims beyond its support.

4. Check design, journal and governance requirements

   Select the reporting checklist appropriate to the design, locate its items in the manuscript and check current journal author instructions. Author contributions, ethics approval or exemption, informed consent, registration, data-use permission, conflicts of interest and funding must be entered by responsible people from real records. Unknown facts remain pending; software must not fill them in.

Eight checks before submitting files

Keep a location or evidence record for each item, not just a tick:

• Are the main question, design, population, time frame and prespecified analysis consistent?
• Can eligibility, missingness, sample counts, event counts and all denominators be reproduced or traced?
• Are main estimates and units consistent across abstract, text, tables and figures?
• Are table titles, footnotes, captions, error bars, abbreviations and schematic labels clear?
• Do key citation identifiers, original text and corresponding claims match individually?
• Do limitations cover design, measurement, missingness and causal-interpretation boundaries?
• Have applicable reporting checklists and current journal author instructions been checked?
• Have authorship, ethics, registration, permission, funding and conflict-of-interest facts been confirmed by responsible people?

If the facts in the last item remain unresolved, preserve the gaps and pause submission. No automated checking result constitutes journal permission or a promise of acceptance.

Frequently asked questions