Scientific Manuscript

5 Easy Steps for Writing a Scientific Manuscript of a Population-Based Study

I often receive questions about how to write a successful, publication-worthy scientific manuscript. In 5 easy-to-follow steps, you can develop and polish your written work into a product worthy of publication:

  1. Develop your tables
  2. Start writing the results section first
  3. Write a succinct introduction that justifies the study
  4. Summarize your measures and approach in the methods
  5. Elicit a thoughtful discussion  

1. Develop your tables

Assuming you have a clear and testable overarching research question and analyses performed, you can start working on the tables. The truth is that the tables should keep you honest about addressing your specific research question.

The first table should generally describe your study population. I generally like to include here distributions of population characteristics in the overall sample. Authors often show these distributions stratified on the outcome variable to determine whether any of the characteristics are strong predictors of the outcome. If they are, then these characteristics can contribute to confounding, particularly if they are also associated with the exposure. Careful adjustment for these characteristics is warranted.

The next tables should answer your research hypotheses more explicitly. Devote one table for each hypothesis, whether it involves a subgroup or secondary data analysis. Hypotheses that test for effect modification (sometimes referred to as interaction or moderation) should be indicated as exploratory, since they are difficult to find—often, a large sample size is necessary and, if you decide to test multiple variables for interaction, it’s likely that you’ll find one purely by chance.

Here are some helpful tips on writing your tables:

  • Include clear and succinct titles that describe how the data are presented.
  • Provide sufficient information on the tables so that they stand alone and won’t require the reader to refer to the text.
  • Use as many footnotes as is necessary to describe important information about how the variables were treated in the analysis (e.g., label cut-points, describe any missing data, show all adjustment variables and how they were entered in statistical models).
  • Avoid presenting overly lengthy p-values, particularly when you have very wide confidence intervals. Report them to 2 digits beyond any leading zeros.
  • Include any statistical tests used to derive your p-values. Describe them in the footnotes.

What about figures?

Figures can easily be produced from a set of data points that are readily available. I tend to have some ready-made tables—more than what I generally need for the manuscript. Some of these tables, particularly those that are peripheral to my hypotheses, can serve as supplemental material. Others can serve as figures.

Sometimes the best way to tell an important part of your story is through figures. For example, a flowchart can seamlessly show the reader how you arrived at your final analytic sample. However, most journals usually have a limit to how many tables and figures you can submit for publication so be judicious. 

2. Start writing the results section first

It doesn’t seem intuitive at first but there is a strategic advantage in drafting the results section as the next step. First, it forces you to finalize your tables—and by extension, your analysis— since the text of the results section should flow logically and mirror what’s presented on the tables. Second, it helps you to identify where the main focus of your manuscript will be. You’ll therefore have a better sense of what the introduction and discussion sections will look like.

It’s generally a good idea to start by writing a paragraph for each of your tables, summarizing the main findings. For example, table 1 might describe the distributions of population characteristics in the overall study population and whether these distributions differ when stratified by outcome and/or exposure status.

More helpful tips on writing your results:

  • Be sure not to repeat every result from your tables or figures. Describe only the main findings.
  • Show confidence intervals when reporting parameter estimates and, wherever possible, avoid using the word “significant”. An estimate with a p-value of >0.05, while not statistically significant, can still be important and meaningful.1
  • Remember not to imply causation unless the study is randomized. Avoid using the word “effect” when describing an “association”.
  • Be cautious in your interpretations of parameter estimates. Overly wide confidence intervals can render these values as imprecise.
  • Avoid any editorial comments. Leave this for the discussion section.

3. Write a succinct introduction that justifies the study

The purpose of the introduction is not merely to provide the background material for your topic. You should aim to provide a compelling rationale for why you’ve conducted the study. This can be successfully executed with a few paragraphs that contain the following elements:

  • Clearly and succinctly state the problem that you are addressing.
  • Discuss what is known (and unknown) about the problem and the population under study.
  • If applicable, introduce the theoretical framework or approach that you will use to address the problem.
  • Include a few sentences towards the end to describe the research question(s) that you are addressing. This is an important step that often gets missed. Your research question(s) should be clearly stated.

What about length?

I tend to prefer short (1-1.5 pages) introductions, and so do most epidemiology and clinically-oriented journals. However, social science journals allow for longer introductions. If you are not sure which journal you will submit to, start drafting a long (2-3 page) introduction and cut out material later if needed. It’s a lot easier to shorten text than it is to add it. 

4. Summarize your measures and approach in the methods section

The “methods and material” section is a place to describe all the technical aspects of your study. It’s helpful to divide this section into 3 main parts:

  • Study population. Here is where you should document how the participants were sampled, recruited, and/or retained (if a longitudinal study was undertaken). Mention who were excluded and why. Be sure to list the final analytic sample size; i.e., the sample you are running the statistics on.
  • The variables that you are using for the study were collected somehow. Here is where you should describe in a comprehensive fashion how the outcomes, exposures and covariates were ascertained. Be specific about the collection process. Was it a semi-structured, validated scale that was used? Did you use a clinical assessment and was it calibrated across multiple raters? 
  • Statistical analyses. When describing the statistical approach, it’s useful to follow the sequence of the tables. In other words, describe the methodology that you used to put together the tables and figures. Provide citations for any novel statistical approaches and reasons why they were applicable in your study. Be specific about how regression models were constructed and how you treated variables in the analyses. Specify whether you included any exploratory, secondary, and sensitivity analyses. Indicate whether P values were 1 or 2 sided. It’s also useful to provide the statistical software program that was used.

Your methods section should be comprehensive enough to cover all of your techniques and approaches used in your study. But how detailed should you get?

It’s easy to get carried away trying to describe every technical detail of your study, especially those with complex methodologies. Be mindful that shorter manuscripts are preferred—you don’t want to lose your readers’ attentions with unnecessarily wordy details. Rather than writing them yourself, it’s usually best to provide citations of methodological papers that cover them at greater depth.

5. Elicit a thoughtful discussion

Structure your discussion as an extension—not as a repetition—of the results section. The purpose is to put your findings into broader context and see how they measure up. To elicit a thoughtful discussion, I find it useful to frame this section using concepts of causation and causal inference in mind.2 What you want is to refer back to your research hypotheses, then make your arguments for how well they were addressed and relate them to the broader scientific literature.

Structure your discussion section into 5 parts:

  • Summary of main findings. Start the discussion with a brief summary of the main findings. Was your primary hypothesis confirmed of refuted? How strong was the magnitude of association? Be humble in your writing. Even with strong and meaningful findings, humility can go a long way.
  • Compare to other studies. Gather a list of the most relevant and recent papers you can find to compare your findings, and pay particular attention to the kind of populations that were studied, the strengths of association found, and how your study fits in with the larger story. When describing findings of a controversial issue, be sure to cite papers that cover both sides of the issue. 
  • Offer mechanistic pathways. If you have found an association in your study, it is important to think about whether it can be explained mechanistically. If the extant literature does not offer a suggested pathway, it is perfectly fine to speculate on one that is consistent with your data. But be careful not to overly speculate! Base any speculation on previous evidence. 
  • Don’t forget limitations. No study is perfect. You should dedicate one paragraph to a comprehensive discussion of all the limitations that might have directly or indirectly influenced the results that you found. It’s a good idea to discuss in what direction (towards or away from the null hypothesis) each limitation may have biased your findings. It’s even better if you could provide some evidence to support your speculations. 
  • End with a short conclusion. Offer a take-home message that readers can walk away with. Avoid general statements that suggest “more research is needed”. Rather, be specific about how such future studies might address the gaps of knowledge that remain. Also, highlight the key strengths of your study so you end with a high note.

A final suggestion

Remember to use verbiage that is clear and easy to follow throughout your manuscript. You’d be surprised how many reviewers get confused simply because authors try to convey so many ideas into a very limited space.

As a reviewer myself, I can share that I rarely reject manuscripts that have a single fatal flaw. It’s usually the aggregate of many minor flaws that does it for me. But by following these steps, you can increase the likelihood that your manuscript will be reviewed positively!

 

  1. Greenland S, Senn SJ, Rothman KJ, et al. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. Eur J Epidemiol. 2016;31(4):337-350.
  2. Rothman KJ, Greenland S. Causation and causal inference in epidemiology. Am J Public Health. 2005;95 Suppl 1:S144-150.