EditverseWriting the discussion chapter of a medical thesis
A paragraph plan for the discussion chapter: principal findings, a comparison table against earlier studies, mechanisms, limitations tied to the design, and implications.
Key points
- Open with the principal findings stated against the objectives, in two or three sentences, without repeating the results tables.
- Compare with earlier studies through a table first, then explain the differences in prose: population, definitions, measurement and follow-up.
- Tie each limitation to the design and say which way it is likely to have moved the estimate, and by how much.
- Match the strength of each claim to the design and the confidence interval. Separate clinical importance from statistical significance.
- End with implications for practice and specific next studies, then conclusions that answer the objectives and go no further.
What the chapter has to do
The results chapter says what was found. The discussion says what it means, how far it can be trusted and where it sits in the existing evidence. Examiners use it to judge whether the candidate understands the study's own weaknesses, which is why a discussion that only restates results and lists agreeing papers reads as unfinished.
Two sources give a structure that works for any clinical design. The ICMJE Recommendations, updated January 2026, advise beginning with a brief summary of the main findings, exploring possible mechanisms or explanations, placing the findings in the context of the totality of relevant evidence, stating the limitations, and exploring implications for future research and for practice or policy.1 The STROBE explanation and elaboration paper recommends that authors structure the discussion, perhaps with subheadings, and its four discussion items (key results, limitations, interpretation, generalisability) are a ready checklist for observational theses.2 The case for a structured discussion was made in a 1999 editorial in The BMJ.5 For randomised trials, CONSORT 2025 has two discussion items: interpretation consistent with the results, balancing benefits and harms and considering other evidence; and limitations covering bias, imprecision, generalisability and, where relevant, multiplicity.4
Your university may prescribe headings or a length for the chapter. Follow its format, and fit the plan below inside it.
Paragraph plan
| Block | What it contains | A sentence that starts it |
|---|---|---|
| 1. Principal findings | The answer to the primary objective with its estimate and interval, then the main secondary findings. One paragraph. No new numbers. | "This study set out to [objective]. We found that [finding], with [estimate and 95% CI]." |
| 2. Comparison with earlier studies | A comparison table, then paragraphs grouped by finding, not by paper. Explain agreement and disagreement. | "The infection rate in our standard-care group was higher than in [study], which may reflect [difference]." |
| 3. Mechanisms and explanations | Biological or practical reasons the finding is plausible, with references. Marked as explanation, not as something the thesis showed. | "One explanation for this association is [mechanism], which is supported by [evidence]." |
| 4. Strengths | Features of design or conduct that support the result: prospective data, defined outcome, complete follow-up. Brief. | "The outcome was assessed with a prespecified definition by an assessor unaware of group." |
| 5. Limitations | Each source of bias or imprecision, its likely direction and size, and what was done about it. | "Because [limitation], the true [estimate] is likely to be [higher or lower] than we observed." |
| 6. Generalisability | The patients and settings to which the result applies, and those to which it may not. | "These findings apply to [population] treated in [setting]; they may not hold for [group]." |
| 7. Implications and conclusion | What a clinician might do differently, what study should come next, and a conclusion tied to the objectives. | "A randomised comparison with [feature] is needed to confirm [finding]." |
Blocks 5 and 7 belong together. STROBE's authors ask that research recommendations be linked to the limitations and that investigators suggest how later work can improve on theirs instead of stating that more research is needed.2 If the main limitation was a single centre and a short follow-up, the recommendation is a multicentre study with longer follow-up, named as such.
The comparison table
Build this table before writing block 2. It forces a like-for-like reading of earlier work and often shows that apparent disagreement is a difference in definition. The entries below are invented placeholders that show the layout. They are not real studies and must not be cited.
| Study | Design and setting | Participants | Outcome definition and follow-up | Main result | Difference from the present study |
|---|---|---|---|---|---|
| Present study | Prospective cohort, one teaching hospital | 194 | Infection by protocol definition, 30 days | 9.4% vs 19.4%; risk difference −10.0 points (95% CI −19.8 to −0.3) | Reference |
| Study A (placeholder) | Randomised trial, three centres | 420 | Infection by a surveillance definition, 30 days | 7.1% vs 12.9% | Elective cases only; lower baseline risk |
| Study B (placeholder) | Retrospective cohort, one centre | 260 | Infection recorded in notes, until discharge | 6.2% vs 8.5% | No post-discharge follow-up; infections after discharge missed |
| Study C (placeholder) | Randomised trial, one centre | 150 | Infection by protocol definition, 14 days | 10.7% vs 21.3% | Emergency surgery included; shorter follow-up |
The prose that follows the table explains the pattern. A model paragraph for this illustrative table:
"The infection rate with standard dressings in our cohort (19.4%) was higher than in Study A (12.9%) and Study B (8.5%). Two differences are likely to account for this. Our cohort included emergency operations, which Study A excluded, and we followed patients for 30 days, whereas Study B counted only infections recorded before discharge. Study C, which also included emergency cases, reported a rate close to ours. The direction of the difference between dressings was the same in all four studies, although our estimate is imprecise and its interval includes differences too small to matter clinically."
Notice what the paragraph does not do. It does not list each study in turn with "similar to our study" or "in contrast to our study", and it does not treat agreement as proof.
Limitations tied to the design
STROBE item 19 asks authors to discuss limitations taking into account sources of potential bias or imprecision, and to discuss both the direction and the magnitude of any potential bias.2,3 A limitation named without its consequence does not meet that standard. Start from the design.
| Design feature | Typical limitation | How to write it |
|---|---|---|
| Retrospective record review | Incomplete or inconsistent recording | State which variables were missing and for how many. Say whether events are more likely to be missed in one group, and so which way the comparison is pushed. |
| Non-randomised comparison | Confounding by indication | Name the factors that influenced who received which treatment, which were adjusted for, and which could not be measured. |
| Cross-sectional data | No temporal order | Say that exposure and outcome were measured together, so the association cannot show which came first. |
| Single centre | Limited generalisability | Describe how your patients and practice differ from other settings, and what that implies for applying the result. |
| Small sample | Imprecision | Point to the width of the confidence interval and say what range of true effects is compatible with the data. |
| Unblinded outcome assessment | Detection bias | Say whether the assessor knew the group and whether the outcome involved judgement. State the likely direction. |
| Loss to follow-up | Selection bias | Give numbers lost by group and compare their baseline characteristics with those who remained. |
| Many secondary analyses | Multiplicity | State how many comparisons were made and treat secondary and subgroup findings as exploratory. |
Compare two versions of the same limitation. Weak: "The sample size was small." Stronger: "With 28 infections in total, the confidence interval for the risk difference was wide (−19.8 to −0.3 percentage points). The data are compatible with a large benefit and with almost none, so the study cannot establish the size of the effect." The second version tells the examiner that the candidate can read their own interval.
Wording the interpretation
The ICMJE guidance on conclusions is specific: link them to the goals of the study, avoid unqualified statements and conclusions not adequately supported by the data, distinguish clinical from statistical significance, avoid statements on economic benefit unless the study includes economic data, avoid claiming priority, and label new hypotheses as such.1 It also asks authors to discuss the influence of variables such as sex or gender on the findings where appropriate; the Recommendations cite the SAGER guidelines for reporting sex and gender.8
| Phrase to avoid | Why | Write instead |
|---|---|---|
| "This study proves that..." | A single thesis study does not prove | "These findings are consistent with..." or "This study provides evidence that..." |
| "X causes Y" from an observational study | Association only | "X was associated with Y after adjustment for..." |
| "There was no difference" | A non-significant result is not evidence of no effect | "We did not detect a difference; the interval includes effects from A to B." |
| "Highly significant" | A P value does not measure the size or importance of an effect | Give the estimate, its interval and whether the size matters clinically. |
| "A trend towards benefit" | Presents a non-significant result as positive | Report the estimate and interval and state the uncertainty. |
| "This is the first study to..." | Priority claims are hard to verify | "We found no earlier study in this population in our search of [databases, dates]." |
The term for presentation that makes results look more favourable than the data support is spin. In a review of 72 reports of randomised trials with non-significant primary outcomes, the reviewers identified spin in the discussion section of 43% of reports.6 The commonest misreadings of P values and confidence intervals, with correct wording for each, are catalogued in a 2016 guide in the European Journal of Epidemiology and are worth checking against your own interpretation paragraphs.7
Before the chapter goes to your guide
- The first paragraph answers the primary objective and could be read alone.
- Every number in the discussion already appears in the results chapter and matches it.
- The comparison table has a row for the present study and the same columns for every study.
- Each difference from earlier work has a proposed reason.
- Each limitation has a stated direction and, where possible, a size.
- Subgroup and secondary findings are described as exploratory.
- Recommendations name a specific design, population or outcome.
- There is one conclusion for each objective, and none goes beyond the data.
- No results are introduced for the first time in this chapter.
Where an editor helps
The interpretation is the candidate's own and is what the viva tests. An editor's contribution to this chapter is to check that it is complete, consistent with the results and clearly written. Under MD, MS and master's thesis editing, a subject-matter editor returns tracked and clean files formatted to the university template with a reference check. The higher tiers add developmental comments and an editor's report, which is where gaps such as a limitation without a stated consequence, or a claim stronger than the interval allows, are flagged for the candidate to resolve. Where the difficulty is reading the statistical output itself, statistical results interpretation provides a statistician's written explanation of what each result does and does not show. Both follow the candidate's university rules. See medical dissertations for the full list.
Frequently asked questions
Should I repeat the results in the discussion?
Summarise the main findings briefly at the start. The ICMJE guidance is not to repeat in detail data given in other parts of the manuscript. Quote a number again only when the argument needs it.
How long should the discussion chapter be?
Your university sets any limit. Within it, length follows content: one paragraph for principal findings, as many comparison paragraphs as there are findings to compare, and a limitations section that covers every design weakness.
Do limitations weaken the thesis?
No. STROBE describes the discussion of limitations as an essential part of scientific reporting. A limitation with its direction and size stated shows the examiner that the result has been tested by its own author.
What is the difference between the discussion and the conclusion?
The discussion interprets and qualifies. The conclusion states, in a few sentences linked to the objectives, what the study supports. Recommendations follow from the limitations and the conclusion.
References
- International Committee of Medical Journal Editors. Preparing a manuscript for submission to a medical journal. In: Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals. Updated January 2026. icmje.org
- Vandenbroucke JP, von Elm E, Altman DG, Gøtzsche PC, Mulrow CD, Pocock SJ, et al. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration. PLoS Med. 2007;4(10):e297. PMID 17941715 · DOI
- von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet. 2007;370(9596):1453-7. PMID 18064739 · DOI
- Hopewell S, Chan AW, Collins GS, Hróbjartsson A, Moher D, Schulz KF, et al. CONSORT 2025 statement: updated guideline for reporting randomised trials. BMJ. 2025;389:e081123. PMID 40228833 · DOI
- Docherty M, Smith R. The case for structuring the discussion of scientific papers. BMJ. 1999;318(7193):1224-5. PMID 10231230 · DOI
- Boutron I, Dutton S, Ravaud P, Altman DG. Reporting and interpretation of randomized controlled trials with statistically nonsignificant results for primary outcomes. JAMA. 2010;303(20):2058-64. PMID 20501928 · DOI
- Greenland S, Senn SJ, Rothman KJ, Carlin JB, Poole C, Goodman SN, et al. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. Eur J Epidemiol. 2016;31(4):337-50. PMID 27209009 · DOI
- Heidari S, Babor TF, De Castro P, Tort S, Curno M. Sex and Gender Equity in Research: rationale for the SAGER guidelines and recommended use. Res Integr Peer Rev. 2016;1:2. PMID 29451543 · DOI