The discussion
The discussion of a research paper explains what its results mean: it opens by answering the question the study asked, compares the findings with the literature, states the strengths and limitations, and closes on the conclusions and their implications.
What the discussion is for
The discussion should answer the question "So what?" by stating the relationships among the facts discovered, relating them to prior studies, including those mentioned in the introduction, and postulating what it may all mean: the conclusions. Its main purpose is to explain the implications of the results.
The discussion is an opportunity to contextualise the findings, to show how they support or refute the hypothesis being tested, and to show how they lead to new hypotheses or research. Its contents should flow directly from the results and focus mainly on the findings observed, which can include citing other literature on those findings and suggesting possible explanations for them.
In the hourglass shape of a manuscript, the discussion widens again from the specifics of the objectives and methods to the broader clinical relevance of the findings, presenting their significance and their implications for the field. It is key for readers who want to put the study's implications into perspective, and it should highlight both the newness and the importance of the findings. Readers often look ahead to the discussion for a summary of the findings, their relevance for the field and their clinical implications, so the discussion should be both concise and objective.
In a clinical manuscript, whether it reports a drug trial, the natural history of a disease or a new diagnostic approach, the discussion is where the study's results should be placed in a context meaningful to the journal's readers, such as clinicians and other researchers.
Many writers and authors find the discussion the most challenging section of a manuscript to write. Writing a good discussion is one of the biggest challenges for many authors. One experienced teacher of writing holds that it is the hardest section to write and to get right.
The order of its parts
The beginning of the discussion should answer the research question. The middle should explain how the results support that answer, and the end should restate the answer.
A simple structure for the discussion includes five parts: restating the main findings; a brief discussion of the literature on those findings and of how they advance understanding of the hypothesis; a possible explanation for the findings based on biological mechanisms or methodology; the strengths and limitations; and the next steps and overall conclusion.
The discussion can be in any order, and is perhaps less rigorous than other sections of the paper, but most journals tend to follow the same pattern: the key findings, with no new data; how they relate to what is happening in the world, and how the data compare with others'; the strengths and weaknesses of the paper; and any differences from other publications, and their strengths and weaknesses. The pattern then asks what the findings mean for the readers the paper is for, such as whether clinicians can now consider treating a particular patient or whether the work is another step and further research needs to be done, and whether there should be further research and whether the work opens new questions. The pattern can be used as a checklist, starting each item with a keyword and building it up into a phrase.
Making the discussion flow logically from one element to the next, so that it tells a story, can be difficult. A recommended flow is a good place to start, and can be adapted to the writer's own needs and writing style.
A template for the discussion of a clinical manuscript has four major areas, addressed in this order or with the middle two reversed: a conceptual summary of the study; "flies in the ointment", meaning the study's limitations; comparison to the current literature; and the bottom line.
In the conceptual summary, within the first one or two paragraphs, the authors should restate the study's main findings, which puts them in a position to place those findings in context through the remaining areas.
The limitations area rests on two facts: clinical research does not always reveal what researchers expect, and no study is perfect. Drug trials, for example, generally include patients selected against a strict set of inclusion and exclusion criteria, so a trial may not be generalisable to patients unlike its population. For these reasons, the next paragraphs should address these questions:
- What are the limitations of the study, with regard to design, implementation, analysis, interpretation and clinical application?
- Were there unexpected or unfavourable findings?
- Are there explanations for these findings?
- Are these findings considered to be related to the drug, device or other experimental treatment in the study?
The introduction will have summarised the current state of knowledge and the reason for conducting the study, and in the comparison the authors should draw on an extensive knowledge of the current literature to place the new results in a meaningful context for readers. Questions pertinent to the comparison include:
- Are the results consistent with those of previous publications?
- If the results are not consistent, can the differences be explained?
- What do the new results add to the current state of knowledge?
The bottom line addresses the final questions, "So what?" and "Who cares?". The authors should conclude with a clear, concise explanation of why the research matters to patients and to the field and, where appropriate, indicate what questions remain to be answered. The final paragraph or paragraphs should consider:
- How does the study change current clinical knowledge?
- What are the implications for patients? Which patients?
- Should the results change the current treatment or diagnostic paradigm?
- Where do we go from here?
The template breaks the sometimes daunting process of writing a clinical discussion into smaller, more manageable portions, and focuses attention on the information most relevant to readers. A writing team can use it to generate discussion and debate about the study findings and their clinical implications, and it is useful both as a coaching tool for new writers and as a reminder for experienced ones. It provides a framework for evaluating the clinical relevance of new findings in an accurate, balanced and objective way, so that they may be of greatest benefit to patients and future researchers.
What guidelines and journals ask for
The ICMJE recommendations ask authors to emphasise the new and important aspects of the study and the conclusions that follow from them in the context of the best available evidence, without repeating in detail data or other information given elsewhere in the manuscript. They recommend that the discussion briefly summarise the main findings, then explore possible mechanisms or explanations for them, compare and contrast the results with other relevant studies, state the limitations of the study, and explore the implications of the findings for future research and for clinical practice. They recommend linking the conclusions with the goals of the study while avoiding unqualified statements and conclusions not adequately supported by the data. In particular, they recommend distinguishing between clinical and statistical significance, and avoiding statements on economic benefits and costs unless the manuscript includes the appropriate economic data and analyses. In an epidemiological study of 4 million people, for example, an odds ratio of 1.05 could easily be statistically significant without necessarily being clinically meaningful. They also recommend avoiding allusion to work that has not been completed, and stating new hypotheses when warranted.
The CONSORT 2010 Checklist has three items for the discussion:
- Item 20: trial limitations, addressing sources of potential bias, imprecision and, if relevant, multiplicity of analyses.
- Item 21: generalisability (external validity, applicability) of the trial findings.
- Item 22: interpretation consistent with the results, balancing benefits and harms, and considering other relevant evidence.
Both the Uniform Requirements for Manuscripts Submitted to Biomedical Journals and the CONSORT statement ask a clinical discussion to summarise the main findings, explore potential explanations for them, compare the results with those already published, state the limitations of the study, including the generalisability of the findings, sources of potential bias and sources of imprecision in the results, and explore the clinical implications of the findings. The four-area template is consistent with both.
The British Journal of Anaesthesia includes in its instructions to authors the elements a discussion should contain and the pattern it should follow: the main finding; its relationship to previous studies; additional, secondary findings; their relationship to previous studies; limitations; strengths; future directions; and the conclusion.
The opening
After answering the question the study sought to answer, the discussion should briefly explain and defend that answer. Opening lines that work well, with their tenses, include "This study shows that…" in the present, "In this study, we have shown that…" in the present perfect, and, perhaps most commonly, "We found that…" in the past. "Our study showed…" is another good way to begin. An overblown claim such as "This is the first report of…" is best avoided, and a quick hedge added instead, as in "To our knowledge, this is the first report of…".
The opening should outline what the study showed and explain how the data collected address the study objectives. Restating the main objectives as the introduction presented them is fine, because it reminds the reader what they were. Highlighting key findings at the start is also fine, but the discussion should not simply repeat all the results: the results section presents and describes them, and the discussion interprets them and discusses their implications. For example, a results sentence reports that child-related stressors were the strongest predictor of membership in the high-stress group, with an odds ratio of 2.16, and the discussion interprets it as showing that childhood stress is one of the strongest predictors of stress in adulthood. Some authors may use the discussion simply to restate the findings, and a discussion written that way lacks depth and omits information that is important to the journal's readers.
Comparison with the literature
The body of the discussion should compare and contrast the study's main take-home points with those of previous studies, whether by the same authors or by others. The comparison should include references to articles by the key people researching in the field. Citing them shows that the authors know the field, and does no harm if those people end up refereeing the manuscript. Whether also to reference opinion leaders is a question of style, and whether to reference reviewers is a personal choice, settled between the writer and the client.
Discrepancies with related studies should be reported, with an attempt to explain them, rather than passed over as though they did not exist. Alternative explanations for the findings should also be offered, which again shows that the authors know their field and have carefully considered the meaning of their results. A detailed literature review on the hypothesis belongs in a review article, because the space available in a discussion generally cannot accommodate one.
The discussion is where the authors might give their opinion of the novelty of their findings, or of how the findings may affirm or contradict previous research or experience in the field. A study of psychotropic medication and violence after release from prison, for example, noted uncertainty about whether treatment for released prisoners with mental disorders should focus on criminogenic rather than mental health-related factors, and stated that its observational findings supported the potential role of treating psychiatric disorders, including with psychotropic medication.
Limitations, strengths and generalisability
The discussion should explain the study's strengths as well as its limitations. For example, an initial study of 37 patients did not allow formal conclusions about the efficacy of a malaria treatment, but it is highly relevant to understanding malaria treatment because it was performed in a region of high transmission. Where a particular limitation did not affect the results, the discussion should make this clear, and it should explain the steps taken to limit the influence of potential biases and other limitations.
The limitations should be addressed straightforwardly but succinctly. Weaknesses of the study design should be stated, or reviewers or letters to the editor will surely describe them. The discussion should also describe any factors that may have biased the collection of the data, such as unexpected events, attrition of subjects or changes in methods partway through the study, such as terminating one of the study groups. A study of aircraft noise and fatal myocardial infarction, for instance, describes the possibility of bias in the coding of deaths.
The limitations worth stating are the study's true limitations, not the obvious ones. A randomised trial report commonly says that the stringent inclusion and exclusion criteria used to enrol its patients do not warrant generalising the results to the general population, but that more or less states the obvious, because any trial with inclusion and exclusion criteria selects the population it includes. A true limitation could be, for example, a study only three months long in a chronic condition, so that nothing is known of what can happen in the long term, or data the authors believe could be important to understanding the outcomes that could not be collected in the setting where the study was performed.
Comprehensively presenting the study's potential strengths and limitations is pivotal to fair balance in the discussion, and only a fair-balanced, credible manuscript has a favourable prognosis with peer reviewers. The aim of the limitations subsection is to consider the study's strengths and limitations and, if possible, argue why the limitations do not abolish the overall conclusions. Without unduly constraining the conclusions, the discussion should acknowledge whether the limitations may have led to underestimation of treatment effects, a bias toward the null, or to overestimation, a bias away from the null, and what future research is needed to address them. Every set of observations has limitations, but a discussion that spends inordinate space explaining away findings that do not support the investigators' hypotheses is usually unhelpful and distracting.
The ultimate aim is generalisability: whether what the study found can be extended to the population at large, or applies only to the group studied, a weakness of the small sample or the single-institution study. Ideally the findings have implications for similar patients elsewhere, the advantage of a large trial involving thousands of subjects at various sites. Describing generalisability means saying how the results of the study bear on the general population that has the illness studied. A pivotal challenge that peer reviewers often take up is ecological validity: a study typically has at least two sample populations, such as cases and controls, or actively treated subjects and placebo controls or usual-care recipients, and the question is how findings in these populations relate to the overall general population. Generalisability is a relatively young word, but it has become a key concept for clinical studies, and how generalisable a study's results are to the wider population helps to determine how broadly the tested treatment can be applied. Generalisability should be described realistically, without unwarranted claims, which journal editors, referees and readers will not accept.
Many authors are reluctant to discuss the limitations of their work for fear that they imply weaknesses in their results. In that situation, it is useful to ask the questions a peer reviewer would ask, to identify possible weaknesses in the study. Acknowledging them in the discussion can pre-empt peer reviewers' questions by giving the reasons certain better approaches were not possible, potentially saving time and effort after submission.
Limitations and generalisability are the parts of the discussion where a medical writer can help the authors most. The writer can give them feedback on the strengths and possible limitations of what they have done. A true limitation may seem obvious to the investigators because it is part of their daily routine, so the writer's peer interaction with the authors is very important: it makes them more aware of what they have done than they are, because authors love the study they have created.
Interpretation and conclusions
Interpretation is the authors' account of what the results imply for daily clinical practice, and something has to be stated even when the outcome is negative and does not change clinical practice. The authors should lead the reader to understand the implications of their results. It is crucial that they interpret the results and, if appropriate, state the risk-benefit ratio or at least clearly state the benefits and the risks of the results they have presented.
The conclusions need to be consistent with the research design, the data quality and the analytic approach used; it is inappropriate, for example, to present a hypothesis-generating finding as hypothesis testing. In the experience of some writers, striving for clarity and brevity reduces the likelihood of overstating the implications and relevance of a finding that happens to be consistent with a hypothesis, and of excusing or rationalising an inconsistent one.
Authors may wish to emphasise a conclusion that is somewhat speculative or that overemphasises the significance of their results. Medical writers need to discuss the overall relevance and implications of the results with the authors, and make sure the conclusion presented is based on the initial objectives stated in the introduction, which ties the manuscript together from beginning to end.
The close
The discussion should end with the conclusions and recommendations. Unanswered questions and future research can be mentioned, but not as the bare statement that further studies are needed. That sentence states the obvious, since one experienced writer knows of no study that needs no further study after it. Where the authors think future studies are needed, the discussion should say what they should aim to do and how. Sharing the authors' view of the next step helps other researchers working on the same topic to set up new studies and possibly to collaborate. Future research should not be the very last thing mentioned; the discussion should instead leave readers with a punchy statement about the importance and implications of the findings.
The end of the discussion should leave the reader knowing why the study matters. Its closing sentences have been likened to an attorney's wrap-up to a jury or the culmination of a salesperson's pitch.
The last paragraph, where a grazing reader may go straight after the abstract, presents a summary of the conclusions and what the team thinks they mean. It states the implications for others in the field, and perhaps how the findings might translate to patient care in the office or hospital. It should be written very carefully, because it represents the outcome of months of effort.
Transitions and new data
Transition words and phrases, such as therefore, however, thus, conversely, consistent with, in contrast to, surprisingly, interestingly and in addition, should be used judiciously and appropriately. Consecutive sentences should not, for example, both start with however, and not every sentence should begin with a transition. Setting the stage up front is usually helpful, as in "In reviewing the literature, we found no data on…".
The discussion should introduce no new data and should not refer to "data not shown". References to data not shown belong in the results section, and data important enough to bring up in the discussion should be presented in the results.
Disagreement among the authors
Any disagreement among co-authors about the interpretation of the results should be described in the discussion. Reports of research studies are often written by committee, whose members seek consensus on what will appear in print, and perhaps for that reason the final paper does not always include the heartfelt opinions of some researchers on the team and often does not reflect the diversity of the authors' opinions. A survey of contributors to ten research articles published in The Lancet, to which 36 of the 54 contributors responded, found unreported concerns about the studies' weaknesses and disagreements among authors about the findings and their significance. It concluded that structured discussion sections, like structured abstracts, may be one remedy for such suppressed opinions.
A second, independent discussion
The discussion and conclusion sections of research articles are too often subjective, containing over-interpretation of data and spin, the manipulation of language to mislead the reader. To address this, the editors of the British Journal of Anaesthesia invited independent experts, who had not taken part in the research and were given only its methods and results, to write a second discussion for a research article published in the same issue. Everyone is biased, and a second discussion gives a second pair of eyes. The article was a randomised controlled trial reporting a secondary endpoint, and it concluded that in elderly patients having hip fracture surgery with spinal anaesthesia supplemented with propofol sedation, heavier intraoperative sedation was not associated with significant differences in mortality or return to pre-fracture ambulation up to 1 year after surgery. The second discussion was written by two anaesthetists, one of whom was the handling editor of the original article, and a biostatistician.
The two discussions interpreted the main result the same way. The major inferential difference between them concerned the appropriateness of the sample size: the original investigators held that the study was large enough to detect a clinically meaningful reduction in mortality, while the independent discussants inferred that the estimated mortality was too high and the estimated decrease in mortality with the intervention unrealistic, so that with only 200 patients the study was not large enough to address the research question. The discussions also differed in emphasis on existing evidence and contextualisation, and on whether comorbidity should be a major issue for future research, and in many other respects they showed inferential reproducibility.
Replication of methods and results is necessary to demonstrate reproducibility but is not sufficient; consistent interpretation in the discussion is also fundamental. Current deficiencies in discussion sections might limit the inferential reproducibility of scientific research. Lack of contextualisation using systematic reviews, over-interpretation and misinterpretation of results, and insufficient acknowledgement of limitations are common problems in discussions, and these deficiencies can harm the translational process. Proposed solutions, including eliminating or not reading discussions, writing accompanying editorials, and post-publication review and comments, do not work very well. A second discussion written by an independent author with appropriate expertise in research methodology is a new, testable solution that could help probe inferential reproducibility and address some deficiencies of primary discussions.
The feasibility of a second discussion is open to question, because it is unclear who would accept spending time writing a discussion for a study they have not done. Reviewers are best positioned to write one, but their incentive is in question, including whether they would then become authors. Editorials can serve some of the function of a second discussion.
One editor holds that discussion sections are probably the weakest section of a paper and must be improved, and that structuring them, as the British Journal of Anaesthesia proposes, is part of the solution. In 2019 few journals had considered structuring the discussion with a standard format.