How to Write a Discussion Section That Referees Cannot Ignore
The discussion section is where most papers lose their reviewers - through overclaiming, undercontextualising, and limitations sections that read like legal disclaimers. Here is what a well-constructed discussion actually looks like.
In a 2015 analysis of 126 general medicine papers published in high-impact journals, researchers examined the correspondence between what the results actually showed and the language used to describe them in the discussion section. Overclaiming - using language that implied stronger causal inference or greater generalisability than the study design supported - was present in the majority of papers. This is not primarily a problem of dishonesty. It is a problem of convention: researchers learn to write discussion sections from reading other discussion sections, and those sections have normalised a register of confidence that routinely exceeds the evidence.
The Opening Statement: What You Found
The first paragraph of a discussion section should state the main finding clearly and in plain terms, without repeating the numerical results that appeared in the results section. This sounds simple and is frequently done badly. Common failures include: restating the results in full rather than summarising the finding; burying the main result under caveats before stating it at all; and opening with a claim that is slightly stronger than what the results showed. The opening statement sets the interpretive frame for everything that follows. If it overclaims, the rest of the discussion will need to walk back from it - and reviewers will notice the inconsistency.
The main finding should be stated once, clearly, in the past tense ("we found that" or "this trial demonstrated that"), and it should match the result with neither inflation nor deflation.
Contextualising Against the Existing Literature
The core function of a discussion section is to place the new findings in the context of what was already known. This requires a genuine engagement with the prior literature - not a selective citation of studies that agree with your findings, but an honest account of where your results sit in the landscape of existing evidence. If your findings confirm a prior meta-analysis, say so, and describe the degree of consistency. If your findings contradict prior studies, engage with the contradiction directly: could it be explained by differences in population, intervention, comparator, outcome definition, or follow-up duration? Is your study methodologically stronger in ways that would explain the discrepancy, or is the discrepancy genuine and unexplained?
Reviewers who know the field will notice a discussion that ignores inconvenient prior studies. The omission signals either that the authors did not know about those studies - a literature review failure - or that they chose not to engage with them - an intellectual honesty failure. Neither creates confidence.
Limitations That Actually Do Something
The limitations section of most papers serves as a liability disclaimer rather than a genuine engagement with the study's weaknesses. The pattern is familiar: two or three sentences acknowledging limitations that are so generic they could apply to almost any study in the field, followed by reassurances that these limitations "are unlikely to have affected our conclusions." This is not a limitations section - it is the appearance of one.
A limitations section that survives reviewer scrutiny names the specific limitations that actually matter for interpreting this study's results, explains the likely direction and magnitude of their effect on the findings, and is honest about whether those effects could have changed the conclusions. If the study is observational, the residual confounding discussion should identify the specific unmeasured confounders most likely to be important, explain why they were not measured, and estimate whether they would be likely to explain the findings. A limitations section that says "confounding by indication may have been present" without engaging with what that confounding would look like in this specific study is a failure of scientific communication.
Calibrating Language to Evidence Strength
The choice of language in a discussion section communicates the author's degree of epistemic confidence - and should be calibrated to the actual strength of the evidence. A randomised controlled trial with low risk of bias and a pre-specified primary endpoint that is confirmed at high statistical significance justifies "demonstrated" or "established." An observational study with multiple potential confounders and a marginally significant p-value justifies "suggested" or "was consistent with." A single subgroup analysis that was not pre-specified justifies "raised the hypothesis that" at most.
The EQUATOR network reporting guidelines - CONSORT, STROBE, PRISMA, and their extensions - are useful for checking that the claims made in the discussion are supported by the methods described. A study that used an observational design should not use causal language in its discussion without explicit qualification. A study with a composite outcome should not claim benefit on individual components that were not primary endpoints.
Clinical Implications: What They Are and Are Not
The clinical implications paragraph is where discussions most frequently overclaim. "These findings suggest that clinicians should consider X in patients with Y" is appropriate when the evidence is from a well-designed RCT in a representative population with a clinically meaningful outcome. It is not appropriate when the evidence is from a post-hoc subgroup of an observational study with a surrogate endpoint. The difference is not always observed in practice, and reviewers who work in clinical settings will notice when the implied recommendation exceeds the evidence base.
Clinical implications should specify: for which patients (who is this finding applicable to?); what action follows from the finding (what should clinicians do differently?); what evidence gaps remain before the implication is fully supported; and whether the finding adds to existing guidance or contradicts it. Vague phrases like "further research is needed" are not clinical implications - they are acknowledgements that the job is not done.
The Closing Paragraph
The closing paragraph of a discussion section should leave the reader with the two or three things most worth remembering: the main finding, its most important contextual qualifier, and the clearest direction for future research. It should not repeat everything covered earlier, should not introduce new material, and should not make claims stronger than those supported by the body of the discussion. A closing paragraph that accurately summarises what the study contributes and what it does not is a rarer thing than it should be, and it is noticed by reviewers who read dozens of papers a month.
At NousLab, the precision of language in evidence synthesis products is something we treat with the same rigour as the methodology. A finding described with more confidence than the evidence supports is not a rounding error - it is a misrepresentation. See how NousLab structures evidence grading and language standards in systematic reviews.