Scientific Writing for Medical Journals: The Complete Guide

A medical paper is read by editors, reviewers, clinicians and systematic reviewers. Each of them needs to find the design, methods and estimates without guessing.

Guide · 18 min read · Directive Publications Editorial Team

What editors, reviewers and readers need from your paper

A medical paper has several readers, and each reads it differently. The editor scans the title, abstract and cover letter to decide whether it belongs in the journal. Reviewers go to the Methods to decide whether the results can be trusted. Many clinicians stop at the abstract. Later, someone writing a systematic review will try to extract your sample size, primary outcome and effect estimate, and will record your study as unclear if they cannot find them.

Scientific writing for medical journals means serving all of those readers at once. That is less about elegant prose than about completeness and consistency: the same numbers in the abstract, text and tables; methods detailed enough to repeat; and claims no stronger than the design allows.

Use one test on every section: could a careful reader answer their question from what you wrote, without emailing you? If not, the gap will surface in peer review.

Decide three things before you write

Many avoidable rewrites come from decisions made after drafting started. Settle these first.

1. The target journal and article type. Read the journal's instructions for authors before drafting. Note the word limit and how it is counted, the abstract format, the reference style, the number of tables and figures allowed, and the declarations required. If you have not chosen a journal yet, our guide to choosing the right journal without getting scammed covers fit and legitimacy checks.
2. The reporting guideline. Identify your study design and download the matching checklist (see the table below). Use it as an outline while you write, not a form you fill in afterwards.
3. The one-sentence message. Write down, in plain words, what the study found and in whom. For example: "In adults admitted with [condition], [intervention] did not shorten hospital stay compared with usual care." Every section should support that sentence. If co-authors cannot agree on it, resolve that before anyone drafts a paragraph.

IMRaD format: the job of each section

The ICMJE Recommendations, most recently updated in January 2026, describe Introduction, Methods, Results and Discussion as the usual structure for original research, and note that it reflects the process of scientific discovery rather than an arbitrary format. The same document says meta-analyses may need different formats, and case reports, narrative reviews and editorials may be less structured.

Treat the medical manuscript structure as a set of questions, each answered in one place:

Keep material in its own section. Findings that first appear in the Discussion, or methods that surface only in the Results, make a paper harder to check.

Match your study design to a reporting guideline

Reporting guidelines are checklists of the details readers need to judge a study of a particular design. The ICMJE encourages journals to ask authors to follow them, and some journals require the completed checklist at submission. The EQUATOR Network keeps a searchable library of guidelines, including extensions for specific designs and specialties.

Check that you are using the current version. CONSORT and SPIRIT were both updated in 2025: the CONSORT and SPIRIT website describes CONSORT 2025 as a 30-item checklist with a flow diagram, and SPIRIT 2025 as a 34-item checklist. The ICMJE also points authors to the SAGER guidelines for reporting sex and gender.

When you complete a checklist, record the page or section where each item appears. If an item does not apply to your study, say why rather than leaving it blank.

A writing order that saves rewrites

You do not have to write a paper in the order it is read. One sequence that works well for original research:

1. Tables and figures. Build them from the final analysis; they fix the numbers everything else quotes.
2. Methods. Write from the protocol and analysis plan while the details are fresh.
3. Results. Walk through the tables and figures in the order of your objectives.
4. Discussion. Now you know exactly what you are interpreting.
5. Introduction. Easier once you know which background the Discussion relies on.
6. Abstract. Condense the finished text, copying numbers from the final tables.
7. Title. Last, so it describes the paper you actually wrote.

Titles and abstracts

Title

The ICMJE describes the title as a distilled description of the whole article that, together with the abstract, should make electronic retrieval sensitive and specific. Reporting guidelines recommend, and some journals require, that the study design appear in the title. STROBE, for example, asks authors to indicate the design with a commonly used term in the title or abstract. Keep abbreviations out of the title.

A pattern that suits most original research:

[Intervention or exposure] and [outcome] in [population]: a [study design]

• Vague: A study of [intervention] in [condition]
• Specific: [Intervention] versus usual care for [outcome] in adults with [condition]: a randomised controlled trial

Some journals also ask for a short running title, which the ICMJE notes is usually no more than 40 characters, including letters and spaces.

Abstract

The ICMJE says original research, systematic reviews and meta-analyses require structured abstracts, but headings and word limits differ between journals, so use your target journal's format. The abstract should give the context, purpose, basic procedures, main findings with specific effect sizes where possible, and principal conclusions, while noting important limitations and not overinterpreting.

Three practical rules:

• Copy numbers, do not retype them. The ICMJE notes that information in abstracts often differs from the main text. Take every figure from the final tables and check again after each revision.
• Give the estimate, not just the verdict. "Mortality was lower" tells a reader far less than the absolute numbers and an effect estimate with its confidence interval.
• Add registration and data identifiers. The ICMJE recommends that journals publish the trial registration number at the end of the abstract, and trial abstracts should include the items the CONSORT group identifies as essential. If your data are in a public repository, give the repository name and persistent identifier at the end of the abstract too.

The abstract is the only part many readers see and the only substantive part many databases index, so give it extra editing time.

Introduction: known, unknown, question

Keep the Introduction short. The ICMJE asks for the nature of the problem and its significance, the specific objective or hypothesis, and only directly pertinent references. It also says the Introduction should not include data or conclusions from the work being reported.

That usually fits in three paragraphs:

1. What is known. The clinical or scientific problem and why it matters.
2. What is not known. The specific gap your study addresses. Be precise: "few studies have examined this" says little, while "no randomised trials have enrolled patients over 80" can be checked.
3. What you did. The objective or hypothesis, worded so it matches the primary outcome in your Methods.

Methods: write for replication

The ICMJE's guiding principle for Methods is clarity about how and why the study was done, in enough detail that others with access to the data could reproduce the results. It also draws a useful line: Methods should contain information that was available when the protocol was written, and anything learned during the study belongs in Results.

For a clinical study, the Methods usually need:

• Design, setting and dates, including recruitment and follow-up periods.
• Participants: eligibility and exclusion criteria, the source population, and how representative the sample is.
• Sex, gender, race and ethnicity: use sex for biological factors and gender for identity and psychosocial factors, say how each was determined, and explain why race or ethnicity is relevant if you report it.
• Outcomes: primary and secondary outcomes, defined as they were prespecified.
• Interventions and measurements: drugs by generic name, dose and route; equipment with the manufacturer's name; established methods referenced; new or modified methods described with their limitations.
• Ethics: approval by an independent ethics committee or institutional review board and, where applicable, how consent was obtained. Keep the documentation, because editors may ask for it.
• Registration: for clinical trials, the registry and registration number. The ICMJE recommends that editors require registration at or before the time of first patient consent for enrolment, and states that ethics approval does not substitute for it.
• Outside help: any organisation paid to help conduct the research, for example with data collection or management.
• AI used to conduct the study: the tool, version and prompts where applicable, in enough detail to replicate the approach.

Statistics: report estimates, not just P values

The ICMJE asks for statistical methods described in enough detail that a knowledgeable reader with access to the original data could judge whether they were appropriate and verify the results. Build these instructions into your draft:

• Quantify findings with measures of uncertainty such as confidence intervals.
• Do not rely solely on P values, which do not convey effect size or the precision of estimates.
• Name the statistical software and version.
• Define statistical terms, abbreviations and most symbols.
• Distinguish prespecified analyses from exploratory ones, including subgroup analyses.

STROBE adds two items that matter for observational studies and are useful for any design: explain how the study size was arrived at, and explain how missing data were handled.

Here is the difference in a single results sentence, using invented numbers for illustration:

• Weak: The intervention significantly reduced readmissions (P < 0.05).
• Stronger: Readmission within 30 days occurred in 15 of 200 participants (7.5%) in the intervention group and 32 of 200 (16.0%) in the control group (risk difference −8.5 percentage points, 95% CI −14.8 to −2.2).

The second lets a reader judge the size, precision and clinical importance of the effect. The first only says a threshold was crossed.

Results: findings in a logical order

The ICMJE asks for results in a logical sequence with the main or most important findings first. For most clinical studies that means:

1. Participant flow: how many were screened, eligible, included, lost to follow-up and analysed. For trials, CONSORT 2025 includes a flow diagram for this.
2. Baseline characteristics of the groups, often as the first table.
3. The primary outcome, with its estimate and uncertainty.
4. Secondary outcomes: all of those named in the Methods, not only the favourable ones.
5. Additional and exploratory analyses, labelled as such.

Three ICMJE rules apply here. Give absolute numbers alongside percentages, so "35%" becomes "42 of 120 (35%)". Do not repeat all the table data in the text; summarise the most important observations. And report data by demographic variables such as age and sex routinely, unless there is a compelling reason not to, which you should explain.

Keep interpretation out. "Readmission was less frequent in the intervention group" belongs in the Results; "suggesting that early follow-up prevents deterioration" belongs in the Discussion, if the evidence supports it at all.

Tables and figures that stand alone

The ICMJE notes that many figures end up used directly in slide presentations, so each table and figure should make sense without the text. Its guidance reduces to a checklist:

• Number tables and figures in the order they are first cited, and cite every one in the text.
• Give each table a short, self-explanatory title, and put explanations and nonstandard abbreviations in footnotes rather than headings.
• State whether a variability measure is a standard deviation or a standard error.
• Do not present the same data in both a table and a graph.
• Make lettering large enough to stay legible when the figure is reduced, and put titles and explanations in the legend, not on the image.
• Add internal scale markers to photomicrographs and state the staining method.
• Keep the original images behind processed figures, such as blots, because editors may ask to see them.
• Get permission to reproduce any previously published figure, including one from your own earlier paper, unless its licence already allows reuse.

Patient images and case details need extra care: the ICMJE says identifying information such as names, initials or hospital numbers should not be published unless it is essential for scientific purposes and the patient, or a parent or guardian, gives written informed consent for publication. Before adjusting any image, read the journal's policy on acceptable processing.

Discussion and limitations without overclaiming

The ICMJE suggests a structure that works for most papers:

1. Briefly summarise the main findings, without repeating the Results in detail.
2. Explore possible mechanisms or explanations.
3. Place the findings in the context of the totality of relevant evidence, including studies that disagree with yours.
4. State the limitations.
5. Explain the implications for future research and for clinical practice or policy.

A limitations paragraph that helps the reader

STROBE asks authors to discuss sources of potential bias or imprecision, and both the direction and magnitude of any potential bias. That is a good standard for any design. For each important limitation, write:

• What it is, for example outcome data taken from routine records rather than a standardised assessment.
• Which way it could push the result, toward or away from no effect, if you can say.
• How much it might matter, including any sensitivity analysis you ran.
• What it means for interpretation: which conclusions still hold and which need caution.

Claims that invite pushback

• Statistical significance presented as clinical importance. The ICMJE asks authors to distinguish the two.
• Economic claims such as "cost-saving" without economic data and analysis.
• Claims of priority, such as "the first study to show", and references to work not yet completed. The ICMJE advises against both.
• Causal verbs for associations. In an observational study, "was associated with" is usually accurate; "reduced" or "caused" usually is not.
• New hypotheses presented as findings. State them when warranted, but label them as hypotheses.

Tie every conclusion to the study's objectives. A conclusion that answers a question the study did not ask is a warning sign to any reviewer.

References and declarations

References

The ICMJE holds authors responsible for citing accurately and for being able to attest that each reference supports the statement it accompanies. In practice:

• Cite original research rather than reviews where possible, and published articles rather than abstracts.
• Mark preprints clearly as preprints.
• Avoid citing articles from predatory or pseudo-journals.
• Do not cite AI-generated material as a primary source.
• Verify references against PubMed or the original source, and check that none have been retracted, unless you are citing the retraction itself.
• Cite datasets by their persistent identifiers. Our guide to citing data, software and code covers the details.
• Follow the journal's reference style. The ICMJE points to the NLM's Citing Medicine and MEDLINE journal abbreviations, but the journal's own instructions take precedence.

Declarations

Their names and positions vary, so follow the journal's template, but prepare these statements early:

• Funding: the ICMJE recommends listing funding sources separately after the abstract and including only direct support for the work.
• Relationships and conflicts of interest for each author: the ICMJE developed a disclosure form for its member journals and encourages other journals to adopt it, but use whatever format your journal asks for.
• Author contributions, and confirmation that every author meets the journal's authorship criteria.
• Ethics approval and consent, naming the approving body.
• Data sharing: ICMJE member journals require a data sharing statement for reports of clinical trials, and it must say whether deidentified individual participant data will be shared. "Undecided" is not an acceptable answer.
• Acknowledgments: the ICMJE advises editors to require written permission from everyone acknowledged, and says AI writing assistance should be reported here.

Language and style

A few conventions settle most style decisions.

Active or passive voice. Most journals accept both. Active voice ("we recruited 240 participants") is usually shorter and shows who did what; Nature Portfolio's author guidance, for example, says it typically helps readers understand concepts and results. Passive voice ("samples were stored frozen") is reasonable when the actor does not matter. Check the journal's style guide and stay consistent.

Tense. Use the past tense for what you did and found ("mortality was lower"), and the present tense for established knowledge and for what your tables show ("Table 2 shows").

Abbreviations. Use only standard abbreviations, spell each one out at first use, and keep them out of the title, as the ICMJE advises.

Units. The ICMJE recommends metric units, degrees Celsius, and millimetres of mercury for blood pressure unless the journal requires otherwise, with laboratory values in both local and SI units. Journals vary, so check the instructions for authors.

Technical words. Reserve "significant", "random", "normal", "correlation" and "sample" for their statistical meanings, as the ICMJE advises.

Describing people. Use neutral, precise and respectful language, and avoid terms that might stigmatise participants.

Hedging. Match certainty to evidence, but avoid stacking hedges such as "may possibly suggest".

Using AI tools responsibly

The ICMJE acknowledges that many uses of AI can be very helpful, especially for authors writing outside their primary language. Its January 2026 update sets clear expectations for authors:

• AI cannot be an author or be cited as one, because it cannot take responsibility for the accuracy, integrity and originality of the work.
• Disclose use at submission. Describe how you used AI in the cover letter and in the relevant part of the manuscript: writing assistance in the acknowledgments, and data collection, analysis or figure generation in the Methods.
• You remain accountable for the output, including checking that it is not incorrect, incomplete or biased, and that no text or image it produced is plagiarised.
• AI-generated material is not an acceptable primary source for a reference.

The ICMJE also notes that using AI to correct spelling and grammar is generally viewed as acceptable, while using it to generate content must be considered in the context of the work. Nondisclosure may require corrective action and, in some circumstances, may be treated as misconduct.

Practical safeguards:

• Check every reference an AI tool suggests against PubMed or the original paper.
• Write your Results and conclusions from your own tables, not from a tool's summary.
• Do not paste unpublished data or identifiable participant information into a tool without checking how it handles data and what your ethics approval permits.
• Read the target journal's AI policy, because journals differ in what they allow and how they want it reported.

A final self-edit before submission

Put the draft aside for a few days, then read it as a reviewer would.

1. The title names the population, intervention or exposure, outcome and study design
2. Every number in the abstract matches the tables and text
3. The objective in the Introduction matches the primary outcome in the Methods
4. All prespecified outcomes are reported, including unfavourable ones
5. Estimates come with absolute numbers and confidence intervals, not only P values
6. Limitations say which way bias could push the results
7. Conclusions stay within the data and the design
8. The reporting checklist is complete, with page or section numbers
9. References are verified and checked for retractions
10. Funding, conflicts, ethics, registration, data sharing and AI use are declared
11. Word count, abstract format and reference style match the journal's instructions

Then ask a colleague outside the project to read only the abstract and tell you what the study found. If their answer differs from your one-sentence message, revise the abstract before anything else. When the manuscript is ready, our guide on how to write a journal cover letter covers the letter that goes with it.

Where to go from here

Most of scientific writing for medical journals is disciplined reporting: the right guideline, complete methods, honest estimates, and conclusions that stay inside the data. Directive Publications' author resources explain how to prepare, structure and format a manuscript for its journals, and you can submit your manuscript when it is ready.