Give each Discussion paragraph a purpose
ICMJE advises a brief account of the main findings followed by their meaning in the relevant evidence, limitations and implications. For observational studies, also consult STROBE items 18–21 on key results, limitations, interpretation and generalizability.[1][2]
On small screens, scroll the table horizontally to compare every column.
| Role | Question to answer | Common problem |
|---|---|---|
| Main finding | What is the most important answer to the study objective? | Repeating every number in the tables. |
| Interpretation and comparison | What does it mean, and why does it agree or disagree with verified evidence? | Listing citations or presenting an untested mechanism as fact. |
| Limitations | Which selection, measurement or analysis conditions constrain interpretation? | Saying only that the sample was small, without identifying the uncertainty. |
| Implications and conclusion | What can these data support, and what remains to be examined? | Turning an observed association into a treatment effect or universal recommendation. |
A worked revision: from repetition to interpretation
This is a fictional teaching example. A single-centre retrospective cohort included one index appointment for each of 200 adults. Attendance within 30 days was 82/100 in the reminder group and 70/100 in the comparison group. Reminder receipt was not randomized, and baseline differences were not adjusted for. This is not a real study, patient dataset or product-performance example.
- First sentence: answer the question. Identify the association and setting. Calling a study observational does not by itself justify any association claim; check the actual analysis.
- Second sentence: bound the meaning. Do not turn appointment attendance into improved clinical outcomes. Do not add satisfaction or cost savings that the manuscript did not measure.
- Third sentence: explain the constraint. Connect the absence of randomization and adjustment to the limits of causal interpretation. Do not invent the direction or size of confounding.
An opening Discussion paragraph
Wording to improve
Attendance was 82% in the reminder group and 70% in the comparison group. The difference was 12 percentage points. The reminder group had a higher attendance rate.Revision or template
In this single-centre retrospective cohort, reminder receipt was associated with higher attendance at the index appointment within 30 days. This finding identifies an association with appointment attendance, rather than evidence of improved clinical outcomes. Because reminder receipt was not randomized and baseline differences were not adjusted for, the observed difference cannot be attributed to the reminder alone.The difference is 12 percentage points. The revision explains the scope of the attendance outcome and possible confounding without repeating all the Results numbers. It adds no confidence interval, P value, new analysis or unverified mechanism. This paragraph is not a complete Discussion or a substitute for statistical reporting.
Compare study conditions before comparing conclusions
Compare populations, designs, outcome definitions and measurement times, including evidence that does not agree with your findings. If an explanation is not established, label it as a possibility rather than a tested mechanism. Fill the worksheet from sources you have actually read.
On small screens, scroll the table horizontally to compare every column.
| Check | Your study | Verified previous study |
|---|---|---|
| Population, setting and design | [Eligibility and setting] | [Source location and conditions] |
| Outcome and timing | [What was measured and when] | [Same outcome? Different timing?] |
| Estimate and uncertainty | [Analysis and adjustment] | [Estimate, interval and analysis conditions] |
| Supported interpretation | [Claim supported by your data] | [Scope supported by the source and differences] |
Explain what is uncertain and why
STROBE’s limitations item addresses potential bias and imprecision, including their direction and magnitude where possible. If the data cannot establish these, do not manufacture a direction or magnitude.[2]
- Established condition Example: data came from one institution and reminder receipt was not randomized.
- Consequence for interpretation Example: factors other than the reminder may have contributed to the difference between groups.
- Unresolved question Example: these data alone cannot establish attendance in other institutions or the causal effect of reminders.
A paragraph scaffold to adapt
Replace brackets with verified information. Remove unsupported sentences or leave the issue explicitly unresolved. Do not add claims of being first, superior or causally effective merely to make the prose sound stronger.
[Main finding in relation to the research question].
Compared with [verified source], our study differed in [population, design or outcome definition], which may help explain [observed agreement or difference].
[Specific limitation] may have affected [estimate or interpretation]; [what the available data cannot resolve].
These findings support [bounded implication], while [unresolved question] requires further investigation.Review the Discussion alongside the rest of your manuscript
Prepare a Word or PDF manuscript containing Discussion, Methods and Results, then select “Start my manuscript review.” Sign in, open the manuscript and run Health Check. Use Section Critique for a focused question about a paragraph; in the whole-manuscript review, prioritize discrepancies between results and conclusions.
Request to adapt: “Separate repeated results, interpretations beyond verified evidence, and limitations not reflected in the conclusion. Identify supporting locations in Methods and Results. Do not invent facts, numbers or citations. List anything the manuscript cannot establish as a question for the authors.”
- Before revising Read each flagged passage and check that the problem is present. Do not add study procedures or analyses absent from the records.
- After revising Reconcile outcome definitions, numbers and conclusion strength across Abstract, Results and Discussion. AI review does not replace raw-data reanalysis or co-author review.
Frequently asked questions
Should I avoid all numbers in the Discussion?
No. Briefly summarize numbers needed to understand the main finding. Focus on meaning, other evidence and limits rather than repeating every table and Results paragraph.
Must the Discussion have exactly four paragraphs?
No. The four roles are a drafting aid. Combine or divide paragraphs according to the research questions, article type and journal requirements.
How should I discuss a nonsignificant result?
Do not infer no effect or association from the P value alone. Consider the estimate and confidence interval alongside the study design and data limitations, and explain the uncertainty using the actual analysis.
Sources and how we checked them
ICMJE Discussion guidance and the original STROBE checklist were checked on November 8, 2026. The paragraph structure, examples and worksheets are educational suggestions created by Paper Buddy, not official forms.