Medical Journal Writing Style

Reviewed: 2026-10-01. Resolve journal, article type, study design, and stage before applying a numerical limit. A publisher family is not one author contract.

Official author sources

The previous generic 250/300/350/300/325 abstract table mixed article types and unverified rules. Do not copy its numbers into a submission. JAMA's original investigation requirements do not apply to every JAMA Network journal or section.

Abstract structure

Choose the actual journal's headings; the following are writing patterns, not universal required labels:

• NEJM-style research: Background, Methods, Results, Conclusions.
• Lancet-style research: Background, Methods, Findings, Interpretation, Funding.
• JAMA original data: Importance; Objective; Design, Setting, and Participants; Interventions or Exposures; Main Outcomes and Measures; Results; Conclusions and Relevance; applicable trial registration.
• The BMJ research: Objectives, Design, Setting, Participants, Interventions where relevant, Main Outcome Measures, Results, Conclusions, Trial Registration.

Summary boxes such as The BMJ's “What is already known” and “What this study adds” are distinct from the structured abstract. Do not import BMJ Open's template or its separate strengths-and-limitations box into The BMJ without checking.

Use assets/examples/medical_structured_abstract.md (relative to the skill root) for placeholder-based drafting. It intentionally contains no simulated drug trial outcomes that could be mistaken for published clinical evidence.

Match claims to evidence

Write the precise population, intervention/exposure, comparator, outcome, time horizon, and estimand. Randomization supports causal interpretation when its assumptions and execution hold; it does not automatically remove bias from nonadherence, missing outcomes, outcome measurement, or selective reporting. Small trials are not automatically “associational” merely because of size.

For observational analyses, use association language unless a causal estimand, identification assumptions, design, and sensitivity analysis justify a carefully qualified causal interpretation. Do not claim benefit from a case report.

A non-significant result does not establish equivalence or “similar safety.” Describe the interval and which important benefit/harm values remain compatible with it. Equivalence/noninferiority requires the relevant design, margin, and analysis. Avoid stock claims about residual confounding as the main limitation of a randomized intention-to-treat contrast; identify the actual source of bias.

Numerical reporting

• Give group denominators, event counts, follow-up, effect estimates, and uncertainty appropriate to the prespecified analysis.
• For binary outcomes, provide absolute risks/differences alongside relative measures when estimable. A hazard ratio is not a risk ratio.
• State the confidence level and whether intervals/tests are adjusted for multiplicity. A point estimate plus a P value is insufficient.
• Report exact P values where practical and follow the journal's small-value convention. Statistical significance alone does not establish clinical value.
• Give NNT/NNH only when meaningful from an absolute effect at a stated time horizon and baseline-risk context. Do not invert a hazard ratio or a significance threshold to obtain NNT. If the risk-difference interval crosses zero, the reciprocal interval is discontinuous; do not invent a finite range.
• Report prespecified outcomes regardless of direction or significance. Label exploratory and post hoc analyses. Report adverse events systematically.

Manuscript workflow

1. Establish the clinical question and the gap the design addresses.
2. Describe study design, setting, participants, recruitment, intervention or exposure, outcomes, sample-size rationale, and prespecified analysis.
3. Report ethics review/consent or the documented exemption, and registration where required. Never invent an approval number, funding source, or registry ID.
4. Show participant flow and analysis populations, including missing outcomes. Loss to follow-up does not automatically justify excluding randomized people from an intention-to-treat analysis.
5. Report results in the prespecified hierarchy, with uncertainty and harms.
6. Discuss strengths, bias, precision, generalizability, and clinical relevance without extending conclusions beyond the data.

For diagnostic and prediction studies, specify the intended use, reference standard/outcome timing, validation design, calibration, discrimination, and threshold-dependent tradeoffs. A high AUC alone does not show patient benefit.

Reporting guidelines

Use the current guideline and the applicable design extension. Check the journal's requested checklist version rather than silently substituting one.

Reporting completeness does not establish low risk of bias, causal validity, or adequate precision. These require assessment of what was actually done.

Venue-specific AI policy

JAMA's current instructions prohibit AI drafting for Opinion, Letters, and Online Comments, and advise against AI generation or formatting of references (standard reference managers remain permitted). Check the exact requested work against that policy; disclosure alone does not resolve a venue prohibition. For allowed work, follow journal-specific disclosure and human-accountability requirements. Other journals' policies must be checked independently.

Final check

• Verify the exact headings and limits from the target article-type page.
• Match abstract numbers to the manuscript, tables, analysis, and registry.
• Keep study-design language, estimand, population, and time horizon consistent.
• Check flow denominators, missing-data reporting, prespecified outcomes, and harms.
• Include required checklist, data-sharing, funding, authorship, and interest statements.
• Inspect the uploaded preview, not just the local draft.