Writing a Discussion Without Overclaiming

"Our findings clearly demonstrate that a brief mindfulness intervention reduces burnout among intensive care nurses and should be implemented across hospital systems."

That sentence opened the discussion of a manuscript I handled a few years ago. The study was a single-site, non-randomized pre-post comparison with 41 nurses and a six-week follow-up. The burnout scores did go down. Nothing about the design could tell us why they went down, whether they'd stay down, or whether a nurse in a rural hospital in another country would respond the same way.

Count the overreach. "Clearly demonstrate" is causal certainty from a design that can't support causation. "Reduces burnout" quietly upgrades a change on one questionnaire into a change in a clinical state. "Should be implemented across hospital systems" leaps from 41 people to policy. None of the three is a lie. Together they make a claim the data can't carry, and every reviewer on that paper said so in nearly the same words.

The frustrating part is that the study was useful. If the authors had written a discussion that matched their design, I think it would have gone to revision rather than rejection. Overclaiming doesn't only annoy reviewers. It makes them stop trusting the parts of your paper that are fine.

What overclaiming actually looks like on the page

Most authors who overclaim aren't trying to oversell. They're writing under the assumption that hedged language sounds weak, and that editors want impact. Editors do want impact. We just know what it looks like when the impact is borrowed from a study that hasn't been done yet.

The patterns I see most often:

• Causal verbs from associational designs. "X leads to Y," "X improves Y," "X protects against Y" in a cross-sectional survey or a retrospective cohort.

• Silent outcome swapping. The primary outcome was null, so the discussion opens with the one secondary outcome that reached significance.

• Treating "not significant" as "no difference." A p-value of 0.12 with a wide confidence interval is not evidence that two treatments are equivalent.

• Generalizing past the sample. Findings from university students become findings about "adults." Findings from one hospital become findings about "clinical practice."

• Subgroup enthusiasm. An exploratory subgroup with a striking effect gets a full paragraph and a speculative mechanism.

This has a name in the literature: spin. There's a body of meta-research documenting how often trials with non-significant primary outcomes are reported in ways that make the intervention look beneficial anyway. It's worth an afternoon. Start with this PubMed search on spin in trial reporting and read two or three of the studies. Once you've seen spin catalogued, you start spotting it in your own drafts.

A verb ladder you can actually use

I give this table to every student I supervise. Pick the strength of your verb to match the strength of your design, not your enthusiasm.

One caveat. Hedging every sentence is its own failure. A discussion where everything "may possibly suggest" reads like the authors don't believe their own data. Hedge where uncertainty actually lives, and say plainly what you're sure of.

Open with the answer, sized correctly

The first paragraph of your discussion should answer the question your introduction asked. Not restate the results section, not open with a literature recap. Answer the question, with the effect size and the honest caveat built in.

Here's a rewrite of that nurse study opening:

Before: "Our findings clearly demonstrate that a brief mindfulness intervention reduces burnout among intensive care nurses and should be implemented across hospital systems."

After: "In this single-site pre-post study, ICU nurses who completed a six-week mindfulness program reported lower emotional exhaustion scores at follow-up (mean change −4.2 points, 95% CI −6.9 to −1.5). Without a control group, we can't separate the program's effect from seasonal variation, regression to the mean or other changes on the unit, but the size of the change is large enough to justify a controlled trial."

The second version is longer and less exciting. It's also the version a reviewer reads and thinks, "These people know what they have." That reaction buys you goodwill for the next four pages.

A simple template for that paragraph:

1. What you found, in one sentence, on the outcome you said mattered.

2. How big it was, with uncertainty.

3. The single most important constraint on interpreting it.

4. What that means, at the scale your design supports.

Comparing with other studies is not a literature parade

The weakest comparison paragraphs I see go like this: "Our findings are consistent with Smith et al., Jones et al. and Lee et al. However, they differ from Patel et al." Then nothing. The reader learns that other papers exist.

Comparison earns its space when it explains something. Why might your result differ from Patel's? Different population, different dose, different outcome measure, longer follow-up, a design that controlled for something theirs didn't? That's the interesting part, and it's the part most drafts skip.

Before: "Our results differ from those of a previous trial, which found no effect."

After: "A previous trial found no effect on exhaustion scores, but it delivered the program in a single two-hour session, whereas ours ran weekly for six weeks. If the difference reflects dose rather than chance, that's a testable prediction for future work."

The trap of only citing agreement

A first-year master's student I supervised, Lina, brought me a discussion draft where every single comparison was supportive. Eleven citations, all agreeing. I asked whether she'd found any studies that disagreed. She had, three of them, and she'd left them out because she thought they'd "weaken the paper."

I told her what I'll tell you. Reviewers in your field know those three papers. Some of them wrote those papers. Leaving them out doesn't make the disagreement disappear. It makes you look either unaware or evasive, and neither helps. She put them back in, wrote a careful paragraph about why the results diverged, and that paragraph was the one a reviewer singled out as the strongest in the manuscript.

If you're working from a systematic review, the PRISMA statement pushes you to discuss the totality of evidence rather than a curated subset. It's a good discipline even when you aren't writing a review.

Limitations that do real work

Here's where I part ways with a lot of standard advice. Most guidance tells you to "acknowledge your limitations." So authors produce a dutiful list: small sample, single site, self-report, cross-sectional design. Four bullet points, no consequences, and the paper moves on as if nothing was said.

That list is nearly useless to a reader. A limitation only matters if you tell me which direction it pushes your estimate and roughly how much. That's the difference between a confession and an analysis.

Before: "This study has several limitations, including a small sample size and reliance on self-reported data."

After: "Burnout was self-reported, and nurses who chose to complete the program may have been motivated to report improvement. That would bias our estimate upward. The small sample mainly affects precision: our confidence interval is compatible with effects ranging from trivial to substantial."

Notice the two limitations do different things. One threatens validity (bias), one threatens precision (noise). Treating them as the same kind of problem is a common mistake, and reviewers notice.

Some practical rules I follow:

• Fewer, deeper. Three limitations with consequences beat eight without.

• Put the most serious one first, not buried after the trivial ones.

• If you ran a sensitivity analysis that addresses a limitation, say what it showed.

• Don't list strengths as a rebuttal in the same paragraph. "Despite these limitations, our study has several strengths" reads as defensive. Give strengths their own short space or let the methods speak.

Reporting guidelines are explicit about this. CONSORT for trials and STROBE for observational studies both include items on limitations, generalizability and interpretation, and you can find the current versions through the EQUATOR Network. I'd encourage you to open the checklist for your design and read those items as instructions for your discussion, not boxes to tick after you've written it.

Implications at the right scale

"These findings have important implications for clinical practice" is a sentence I'd happily ban. It's almost never true of a single study, and when it is, the authors should say exactly which practice and how.

Match your implication to your evidence. A pilot study's implication is usually that a larger study is worth doing, and what it should look like. An observational study's implication might be a hypothesis for a trial, or a signal for surveillance. A large, well-conducted trial can talk about practice, ideally alongside existing guidelines rather than in place of them.

The same goes for "future research." "Further research is needed" tells me nothing. Tell me what research: what design, what population, what outcome, what you'd do differently. If you can't specify it, cut the sentence.

A conclusion your data can defend

The conclusion is where overclaiming does the most damage, because it's the part people quote. It ends up in the abstract, the press release, the tweet, and eventually in someone else's introduction, stripped of every caveat you carefully wrote three pages earlier.

My test is simple. Read your conclusion aloud, then imagine a skeptical colleague asking, "How do you know?" If the honest answer requires a study you haven't done, rewrite it.

Before: "Mindfulness training is an effective strategy to combat burnout in critical care and should be adopted widely."

After: "A six-week mindfulness program was associated with lower self-reported emotional exhaustion among ICU nurses at one hospital. A randomized trial with longer follow-up is needed before the program can be recommended for wider use."

What I learned from a former student, now on the other side

One of my former PhD students, Daniel, is now faculty and reviews regularly for two journals in his area. He told me last year that the first thing he does with any manuscript is compare the conclusion in the abstract with the conclusion at the end of the discussion, and both against the primary outcome in the methods. If they don't line up, he says, he reads the whole paper with his guard up.

I laughed, because I'd taught him to do that with his own drafts. He'd simply turned it into a reviewing habit. Plenty of reviewers do the same thing, whether or not they'd describe it that way. Make sure those three places tell the same story.

This also matters ethically. Both ICMJE recommendations and COPE guidance treat accurate, non-misleading reporting as part of an author's responsibility, not a stylistic preference. Spin that misleads clinicians or the public isn't just a writing problem.

A pre-submission check for your discussion

Before you send the manuscript, go through this once, slowly:

1. Does the first paragraph answer the question from the end of your introduction, using the prespecified primary outcome?

2. Does every causal verb match a design that can support causation?

3. Are effect sizes and confidence intervals mentioned where you interpret findings, not only in the results?

4. Have you discussed at least one study that disagrees with you, and offered a specific reason for the difference?

5. Does each limitation say which direction it likely pushes your estimate, or whether it affects precision?

6. Is any subgroup or secondary finding clearly labeled as exploratory?

7. Are your implications scaled to your design?

8. Do the abstract conclusion, the final paragraph and the primary outcome tell the same story?

If you find yourself failing item 4 because you're not sure what the contradicting literature even is, that's a search problem, not a writing problem. It's also the step where a lot of context gets lost. You read the dissenting papers months ago, the notes live somewhere else, and the discussion gets written from memory. That's one reason I've started pointing students toward an AI-native workspace like PACR, which keeps the reading, summarizing and drafting in one place so the papers you need are still attached to the argument you're making. If you want to go back and look for the studies that complicate your story before a reviewer does, you can search and explore the literature there and pull those comparisons straight into your draft.

The tools help, but the habit is what matters. Write the discussion you'd want to read as a reviewer: clear about what you found, honest about what you didn't, and specific about what should happen next. Then read your conclusion one more time and ask, "How do I know?"

If the answer is in your data, you're ready. If you want a single place to keep that whole chain from first search to final draft together, you can start a PACR workspace and try it on your next manuscript.