The conclusion of a research paper answers the objective you set out, in wording no stronger than your study design allows. An observational study supports association, not causation; a single case supports possibility, not frequency. State what you found for the stated question, add no new data, and name the exact study that should come next.
A conclusion is short, but every verb in it is a claim that a reviewer can test against your design. This guide treats the conclusion as a claim you calibrate: to the question you asked, to the design you used and to the abstract that will carry it into search results.
It does not cover the whole Discussion. For interpreting findings against the literature, see how to write a Discussion section. For naming the weaknesses of your study, see writing a limitations section.
What the conclusion of a research paper has to do
A research paper conclusion has three jobs and no others. It answers the question the Introduction posed. It states how far that answer can be trusted, given the design. It points to the next step, if one follows from the result.
Where the conclusion sits depends on the journal. In the IMRaD structure it is usually the last paragraph of the Discussion, though some journals ask for a separate heading. The rules below apply in either position.
Answer the objective you stated in the Introduction
Copy your objective sentence from the Introduction and place it beside your draft conclusion. The conclusion should read as a direct reply to it. If the objective was to compare two dressings for wound infection, the conclusion reports that comparison. It should not open with patient satisfaction, even if that result was more striking.
Three mismatches appear when the two sentences sit side by side:
- A different outcome. The conclusion leads with a secondary outcome because the primary outcome was not significant. Report the primary outcome first, whatever it showed.
- A different population. The objective named adults in one setting, and the conclusion speaks about all patients.
- No answer at all. The conclusion summarises methods ("we analysed records from two centres") without saying what was found.
Let the study design set the strength of the wording
Study design sets a ceiling on the verbs you may use. Words such as effective, efficacy and superior need a comparison group. A causal verb such as reduces or prevents needs a design that can support cause, which in practice means randomisation. Our guide to choosing between a case report, case series and cohort study explains why each design has its limit.
The wording ladder below climbs from the weakest design to the strongest. Copy the row for your design and replace the bracketed slots.
| Study design | Wording that fits | Wording that overclaims |
|---|---|---|
| Case report | This case adds [finding] to the reported presentations of [condition]. | [treatment] is a safe option for [condition]. / [finding] is common in [condition]. |
| Case series | Among [n] [consecutive or selected] patients, [outcome] was recorded in [n]. These observations justify a comparative study. | [procedure] improves outcomes in [condition]. / [procedure] is superior to [alternative]. |
| Cross-sectional study | [outcome] was more prevalent among participants with [exposure] at the time of the survey. | [exposure] increases the risk of developing [outcome]. |
| Case-control study | Prior [exposure] was more frequent among cases than controls (odds ratio [value], 95% confidence interval [range]). | [exposure] is responsible for [outcome]. / The incidence of [outcome] is [value]. |
| Cohort study | After adjustment for [confounders], [exposure] remained associated with a higher incidence of [outcome]. | Removing [exposure] would prevent [outcome]. |
| Randomised controlled trial, clear result | In [population], [intervention] reduced [outcome] compared with [comparator]. | [intervention] should replace [comparator] for every patient. |
| Randomised controlled trial, unclear result | The trial did not show a clear difference in [outcome]; the confidence interval ([range]) does not rule out a meaningful effect. | [intervention] has no effect on [outcome]. |
The last row matters because a non-significant result is not evidence of no effect. Read the interval before you write the sentence: if it includes an effect that would matter, say so.
Reporting guidelines also set expectations for the interpretation. Find the checklist that matches your design on the EQUATOR Network and check what it asks the interpretation to contain.
A conclusion rewritten to match its evidence
The study below is invented to illustrate the structure; its numbers are not real data. A single-centre retrospective cohort compared early mobilisation after abdominal surgery with later mobilisation. The primary outcome was postoperative pneumonia; hospital stay was not analysed. The odds ratio, adjusted for age and frailty, was 0.62, with a 95% confidence interval of 0.41 to 0.93.
Before (overclaims): Early mobilisation prevents postoperative pneumonia and shortens hospital stay. It should become standard care after abdominal surgery. Further research is needed.
After (matches the design): In this single-centre retrospective cohort, early mobilisation was associated with lower odds of postoperative pneumonia after adjustment for age and frailty. Because mobilisation was not randomly assigned, patients who were well enough to move early may have differed in ways the adjustment did not capture. A randomised trial in adults after abdominal surgery, comparing mobilisation within 24 hours with usual care and measuring pneumonia over 30 days, would show whether the association is causal.
The rewrite makes four changes:
- Prevents became was associated with, because a retrospective cohort cannot show cause.
- Hospital stay was removed. It was never reported in the Results, so it cannot appear in the conclusion.
- The practice recommendation was removed. One observational study at one centre cannot justify changing standard care.
- The vague call for research became a named design, population, comparison, outcome and follow-up period.
Keep new data and practice changes out of the conclusion
Every statement in the conclusion must trace back to a result already reported. A new subgroup, a new number or an outcome mentioned for the first time tells the reviewer that the Results section is incomplete. Move the analysis to the Methods and Results, or delete it.
Practice recommendations need the same discipline. A statement that clinicians should change what they do usually rests on more than one study. Suitable wording names what your study adds to that evidence: "these findings support testing [intervention] in a trial" rather than "[intervention] should be adopted".
Make the abstract conclusion say the same thing as the main text
The abstract conclusion is indexed, quoted and shared without the paper around it, so some readers see nothing else. Drift appears when the abstract is written early and the main text is revised later, or the reverse.
Compare the two conclusions word by word before submission. Check that they name the same primary outcome, the same population and the same strength of verb. The abstract may be shorter, but it must not be bolder. Our guide to writing a strong abstract covers the rest of the abstract.
Write "further research is needed" so a reader can act on it
"Further research is needed" is true of every study, so on its own it tells the reader nothing. A useful version names the design, the population, the comparison, the outcome and the question the new study would settle. Use this template and fill every slot:
A [design] in [population], comparing [intervention or exposure] with [comparator] and measuring [outcome] over [time period], would show whether [open question from this study].
If you cannot fill the slots, the call for research is not yet specific enough to include.
Check the conclusion before you submit
- Does the first sentence answer the objective stated in the Introduction?
- Is the primary outcome reported before any secondary outcome?
- Does every verb sit on the correct row of the wording ladder for your design?
- Is every number and outcome already in the Results?
- Is there no recommendation to change practice that a single study cannot carry?
- Do the abstract and main-text conclusions match in outcome, population and strength?
- Does any call for further research name a design, population, comparison, outcome and follow-up period?
If you find a conclusion in one of our published articles that claims more than its design supports, tell us how to report a problem and we will look at it.