Academic English revision

Your Epidemiology Discussion Isn't a Language Problem — It's a Restatement

Your reader already read the Results. Stop making them re-read it in a slightly more cautious tone.

Your Results section is 1,200 words. Your Discussion is 1,200 words. They mirror each other paragraph by paragraph, almost like you copy-pasted and swapped a few verbs. Your advisor's comment is one line: 'This is a restatement of your findings with added hedging language.' You stare at it for twenty minutes, open a grammar checker, fix three articles. Send it back. 'Still the same problem.' The problem was never grammar. It was that you weren't discussing anything.

The biggest trap in epidemiology discussions: you think it means 'say the results again, but more tentatively.' It doesn't. Your reader already read the Results. They don't need you to say 'we observed a significant association between X and Y' a second time. What they need is: what does this association mean at the population level? Which studies agree? Which disagree? Why do they disagree — is it a design artifact or a real difference? You are writing interpretation, not translation. Repeating your Results in the past tense with 'may' inserted here and there is recitation, not discussion.

The skeleton of an epidemiology discussion is actually fixed. Five moves. JAMA, BMJ, The Lancet — 90% of epidemiology papers use the same structure. Move one: interpret your core finding. Not repeat the number — explain what that number means for a population. Move two: compare with the literature. Agree with some, explain why you disagree with others. Move three: strength of causal inference. What does your design support? A cohort and a cross-sectional study carry different weights, and your language must reflect that. Move four: limitations. Not 'future studies are needed' — specify which limitation biases your estimate in which direction. Move five: implication. For public health practice, concretely — policy, numbers, populations.

Here's what the fix looks like. ❌ 'Our study found a significant positive association between air pollution exposure and respiratory hospitalization (OR = 1.34, 95% CI: 1.12–1.60). These results are consistent with previous studies.' The first sentence is a Results copy-paste. The second is filler — consistent with which studies? ✅ 'The observed 34% increase in respiratory hospitalization per 10 µg/m³ PM2.5 increment aligns with the meta-analysis by Smith et al. (2023) and the time-series analysis from the Chinese cohort (Wang et al., 2022), suggesting a consistent dose-response relationship across populations with varying baseline pollution levels.' Same finding. The second version tells the reader where it sits on the literature map. That's discussion.

Second pair. ❌ 'Limitations of this study include potential confounding factors and the possibility of selection bias. Future studies with larger sample sizes are needed to confirm these findings.' This is 2010 writing. 'Potential confounding factors' — which ones? If you can't name your own limitations specifically, why should readers trust your conclusions? ✅ 'Residual confounding by socioeconomic status remains a concern, as our questionnaire captured only broad categories. Given the known gradient in respiratory disease risk across income groups, this limitation likely attenuates our observed association, meaning the true effect may be stronger than reported.' Specific confounder, direction of bias, consequence for your estimate. Run this through easydue and it will check whether your hedging intensity matches your study design — a cohort and a cross-sectional study should not sound the same, and this tool calibrates that for you.

Third pair. ❌ 'These findings have important implications for public health policy. Further research is warranted to explore the underlying mechanisms.' This sentence could close any paper in any field, because it communicates nothing. ✅ 'If the dose-response relationship we observed holds in prospective cohorts, current PM2.5 thresholds in our city may be underestimating the population-attributable fraction for pediatric respiratory admissions by 12–18%, warranting a reassessment of the local air quality standard.' This is what an epidemiology discussion is supposed to do: translate a number into policy language.

Remember this: a discussion is not a sequel to the Results. It's a translation — from statistical language into population language, from finding language into decision language. When you're stuck, list your core numbers and your literature map first, then throw the draft at easydue for a structural pass. Your job is to fix the logic chain and calibrate the causal claim. Stop submitting 'further research is needed' as a conclusion. Your advisor doesn't want more research. They want to know what your research actually means.