How to Write the Perfect Abstract for Radiology

HomeRadiologyVol. 305, No. 3 PreviousNext CommunicationsFree AccessFrom the EditorHow to Write the Perfect Abstract for RadiologySarah L. Atzen, David A. BluemkeSarah L. Atzen, David A. BluemkeSarah L. AtzenDavid A. BluemkePublished Online:Aug 9 2022https://doi.org/10.1148/radiol.229012MoreSectionsPDF ToolsImage ViewerAdd to favoritesCiteTrack CitationsPermissionsReprints ShareShare onFacebookTwitterLinked In Writing a great abstract is crucial to writing a paper that will ultimately be accepted for publication. First impressions made by your abstract matter—a lot. Your abstract is the first document that is evaluated to determine whether your submission should be sent to external reviewers. Assuming your abstract (and paper) are appropriate for Radiology, your abstract will be evaluated by multiple reviewers, and later by multiple members of our editorial board. If accepted for publication, your abstract is the public “face” of your article on search engines such as PubMed or Google Scholar. In all cases, your abstract is rapidly assessed by the reader to determine the topic, relevance, and quality of your research. First impressions matter.Writing the abstract is often left to a junior member of the authorship team, with scant critical review by other authors. This is unfortunate since senior authors are also frequent reviewers. Those senior reviewers fully understand the first impressions made by the abstract. A well-written abstract will help reviewers and editors understand your study in an instant, piquing their interest. A poor abstract will taint your research, making reviewers and the editorial board question the quality of your research.What makes an abstract great? Is a “perfect abstract” possible—in only 300 words? After reviewing thousands of abstracts, below are some of our suggestions on how to make your abstract stand out. We start with a brief overview of the abstract format. We then expand on several characteristics of your study that can make your abstract stand out from the crowd. These suggestions are an expansion of our article, “Top 10 Tips for Writing Your Scientific Paper: The Radiology Scientific Style Guide” (1), and are also reiterated in the Table.Top Tips from Radiology for Writing a Perfect Abstract1. Be concise and specific.The word limit for abstracts in Radiology is 300 words. This includes five sections: Background, Purpose, Materials and Methods, Results, and Conclusion. To meet this word count, be concise but use specific, clear writing. Avoid vague wording. For example, mention the specific modality (eg, MRI) instead of general terminology like “images” or “imaging.” Your abstract should be understood without having to read the rest of your paper.2. Expand all abbreviations at first mention and limit their use.Expand all abbreviations at first mention. Limit to two or three commonly understood terms whenever possible. Too many or unusual abbreviations can make your abstract almost impossible to read, even for other experts. Avoid making up your own abbreviations—if it is not in your hospital’s standard list of approved abbreviations, then it is probably an ambiguous term. Finally, do not abbreviate common terms to save on word count. For example, a recent manuscript used the abbreviation “BC” 10 or more times in the abstract. BC has no general meaning; the authors used this as shorthand for “breast cancer.” For further reading, check out the post about abbreviations published in the new Radiology: Behind the Scenes blog (https://pubs.rsna.org/page/radiology/blog/2022/5/ryblog_05202022).3. Avoid jargon.Authors are prone to using jargon, which is such a common problem in scientific writing that Nature published an article titled “Words Matter: Jargon Alienates Readers” (2). Consider using an online tool such as the De-Jargonizer (https://scienceandpublic.com) to analyze the amount of jargon in your abstract.4. Include a Background statement that captures attention.Your Abstract Background should be one or two sentences (at most). It must answer the question of why your study was performed and/or why it is relevant to the field. Typically, this information should also be in your introduction. For example, mention if studies are lacking on the association you are investigating or if your study is a follow-up to a pilot study. Try to frame your background statement from a clinical radiologist’s perspective (eg, Why is this study important to my radiology practice?).5. Include a clear and compelling Purpose statement.Your Purpose should describe the research question you are trying to answer. This should be a brief but precise statement that is essentially the same as your aim at the end of your introduction. Often, authors will include a Purpose that does not fit their conclusion. For example, if your Purpose is to compare the diagnostic performance of two different imaging techniques, then do not mention survival outcomes in your conclusion. Your Purpose determines the direction of your entire study.6. Include recent data, indicating the study dates.We expect to see the beginning and end dates of your study period or participant accrual; for example, “Study participants were enrolled from July 2020 to June 2021.” Recent dates are preferred; data collected 10 years ago may be suspect due to outdated technology. Longer periods of follow-up are desirable (3 years instead of 3 months, depending on the research question). Clearly state the length of your follow-up period(s) in your Abstract Materials and Methods section.7. Emphasize multicenter studies.Multicenter studies, along with larger sample sizes, are more representative of the general population than single-center studies and should be mentioned in your Abstract Materials and Methods section. For multicenter studies, mention the number and type of institutions where your patients or participants were identified (eg, a tertiary care center, academic medical center, community hospital, university teaching hospital). Do two centers really make a “multicenter study?” Technically, yes. Yet reviewers are savvy researchers as well; they are busy and do not like exaggeration. For sample numbers with only a few centers or sites, “tone it down” and state that your study involved two centers.8. Include a clearly stated reference standard and primary/secondary outcomes (as applicable).Reference standard versus index test: The reference standard is an accepted standard for the ground truth. Much imaging research seeks to compare the ground truth to the index test (eg, a CT result). It should be clear what each of these is in your Abstract Materials and Methods section. If Abstract space is available, then also state your secondary aim and outcome. A ground truth reference standard may not always be available. For example, if comparing CT (index test) versus MRI (reference standard) for hemangioma detection, about 5% of hemangiomas will be atypical or too small to characterize at MRI. In these cases, you cannot assess the typical diagnostic performance of the imaging test, such as sensitivity or specificity. Instead, you must convince the reviewer that your reference standard is still relevant. Be careful not to overstate your results.9. Focus on a few main results. Avoid obscuring results with too many numbers.Radiology researchers (and papers in Radiology) are famous for generating huge arrays of results. We find radiologists to be the opposite—readers want the big picture, not every minuscule variation. In your Abstract, focus on about three main results. Make it readable. Provide numerical data and P values for all comparisons but try to have a balance of text and numbers. The results should be interpretable on their own without reference to your paper. Provide the labels and units for your numbers—a surprisingly frequent mistake. Yes, even a statistic like hazard ratio is not understood without stating the reference category. If you do not understand the units, ask your statistician for help.How to Make Your Abstract Stand Out from OthersThe above nine points are essential for rapid understanding of your abstract by reviewers, readers, and editorial boards. Below, we expand on additional points that enhance the quality of your abstract (if they are applicable to your study).Type of Study MattersIndicate the type of study in the first sentence of the Abstract Materials and Methods. Identifying your study as prospective or a randomized controlled trial will make it stand out. The different types of studies are summarized below. These are listed in order of most to least compelling study types for reviewers and editorial boards:Prospective studies and randomized controlled trialsThese studies tend to have fewer inherent limitations (eg, bias or confounding) than retrospective studies. They require researchers to think ahead: inclusion and exclusion criteria are identified ahead of time, with the objective of testing a prespecified hypothesis. Prospective studies are usually registered on ClinicalTrials.gov (or other databases). Institutional review boards approve statistical plans and monitor safety in prospective studies at regular intervals.Secondary analyses of prospective studiesSecondary analyses of prospective studies seek to answer questions that may have not been conceived of prior to the design of the prospective study. For example, a prospective study of coronary CT in the emergency department might be designed to determine the safety of CT to triage chest pain in patients for discharge to home versus admission to the hospital. However, a secondary analysis of that same prospective study may also be useful to determine how many patients had actionable lung findings requiring follow-up imaging. Another example is the Multi-Ethnic Study of Atherosclerosis (MESA), one of the largest prospectively enrolled cohorts in the United States. MESA was originally powered to determine whether coronary artery calcium score provided additional information to predict cardiac events, over and above traditional risk factors (eg, age, sex, blood pressure, cholesterol, family history). However, since its inception more than 20 years ago, secondary analyses of MESA have provided widely useful clinical, laboratory, genetic, and imaging data, resulting in more than 2000 peer-reviewed publications to date.Retrospective studiesMany studies submitted to Radiology are retrospective studies, in which the authors assemble a group of patients through medical records with certain characteristics. As an example, patients may be included if (a) age is between 18 and 80 years, (b) both CT and MRI are performed for pancreatic cancer, and (c) MRI was performed in the last 4 years. However, in this example, MRI performed in patients with pancreatic cancer may be a red flag for reviewers. At most institutions, CT is initially used to evaluate patients with suspected pancreatic cancer; only those with potentially resectable disease proceed to MRI. Thus, selecting patients who had both CT and MRI leads to an atypical set of patients—why did they have an MRI scan? These patients were likely preselected as having less disease than a general population of patients with pancreatic cancer. Alternatively, they might have a frequent allergy to iodine contrast. Although the selection bias in the above example is easily understood, other bias in retrospectively assembling a research sample cannot be clearly identified.Statistical Analysis: The Importance of Adjusting for ConfoundersSummarize your statistical analysis (ie, tests performed, descriptive statistics) in the last sentence of the Abstract Materials and Methods. Key statistical terms like multivariable, adjusted, or independent/control variables will make your study stand out. If there is insufficient space in your Abstract, multivariable analysis is more important to include than univariable analysis. An appropriate statement might be as follows: “After adjustment for age, sex, and clinical variables, CT tumor size was associated with a higher risk of death at 3 years…” (give hazard ratio, P value, etc). For example, let’s say that a researcher wanted to study if mortality from hepatocellular carcinoma was similar in the background of viral hepatitis versus alcoholic cirrhosis. But in their hospital, patients with alcoholic cirrhosis had a mean age of 75 years compared with 55 years for those with viral hepatitis. Since age is the demographic variable most likely to be associated with death, clearly age is a confounding variable for this research question. Confounders can mask or alter real associations between the independent and outcome variables (3). Multivariable analysis allows us to understand the relationship between the key, independent variable (X) and the outcome variable (Y) after adjusting for confounders. Multivariable analysis requires relatively large sample sizes, as discussed next.Large Enough Sample SizeFor human studies, include the number of patients/participants and cancers or events such as deaths (if that is your focus) in the first sentence of Abstract Results. Also, include mean or median age (with standard deviation [SD] or interquartile range [IQR], respectively), and the number of men or women, whichever is greater. For example, for 100 patients of 60 men and 40 women, write 100 patients (mean age, 47 years ± 10 [SD]; 60 men) were evaluated. Give the most succinct description possible to tell the reviewer who and what you studied. For studies using deep learning, the sample sizes for your training, validation, and test sets should be explicitly stated. Failure to state study size may frustrate reviewers and make them wonder if the researchers are concealing study flaws (such as insufficient sample size).Multiple confounders require larger sample sizes.When there are multiple confounders, reviewers understand that relatively large sample sizes and numbers of events or outcomes are necessary to perform multivariable analysis. For example, in a recently submitted manuscript, the research group had 23 deaths in a group of 100 patients. But the research team identified 16 confounders they needed to adjust for in their analysis. A rule of thumb is that one independent variable is allowed for every 10 events or outcomes. In the case of 23 deaths, at most two independent confounders could be adjusted for by the research group (23 deaths/10 = 2.3, or two independent variables). The sample size and the number of events were simply too small to study the research question, and the manuscript was rejected because little new information was able to be provided.Your sample size must have sufficient power to understand the associations that you are seeking.Exceptions are made for rare diseases or genetic disorders, such as arrhythmogenic right ventricular cardiomyopathy, in which a sample size of 50 patients might be considered sufficient. However, when a research paper reports on 50 patients with lung cancer, reviewers who might see 50 such patients in a month will question the effort of the research team. For common diseases, such as Alzheimer disease or coronary artery disease, sample sizes should include larger numbers of patients, potentially in the hundreds, particularly when accounting for covariates.Larger sample sizes are ideal, and reviewers know it.Larger sample sizes increase the power of your study and are more representative of the population you are studying. Reviewers understand that small sample sizes can cause a great deal of uncertainty in your results. For example, consider a sample size of 35 patients with cancer; your research shows a drug complication in 10 of 35 (29%) patients. However, the 95% CIs are wide—from 15% to 46% in this example. Is this a helpful study? Maybe. You are reporting that almost half of all patients (46%) get the complication, but is that the right answer? Or is the answer closer to 15%? Knowledgeable reviewers will determine if your result is relevant to their clinical practice.For many research studies, a small sample size, especially for retrospective studies, is the main complaint of reviewers. More leniency is given to prospective studies, as those studies usually employ hypothesis testing with sample size estimates to determine study power. And yes, editorial boards understand that remarkable machine learning tools can be deployed to improve the stability of statistical tests. Still, we scratch our heads about studies that have 3000 radiomics parameters (X variables) trying to predict an outcome such as 50 deaths (Y variable) in a study of 200 patients. It takes a skilled researcher indeed to convince reviewers that their data are valid and representative.Balanced Sexes and Diversity of Race and Ethnicity Are ImportantMen and women may have different risks and outcomes for various diseases. Thus, having a balance of both sexes in your study sample is important, unless evaluating sex-specific diseases such as breast cancer or prostate cancer. We recently received a study regarding CT of lung cancer that included only men. However, lung cancer is the fourth most common cancer in women. Half of all lung cancers in 2022 and half of all deaths from lung cancer will occur in women. It appeared that the research team only answer half the question. To reviewers and the editorial board, this indicated tremendous selection bias by the research team, and the paper was rejected. Your study sample should ideally represent the total population, both by sex and by race and ethnicity.We hope these suggestions are helpful to you in writing your next abstract for Radiology. For more information, please see the Radiology Scientific Style Guide (4).References1. Atzen SL, Bluemke DA. Top 10 Tips for Writing Your Scientific Paper: The Radiology Scientific Style Guide. Radiology 2022;304(1):1–2. Link, Google Scholar2. Woolston C. Words matter: jargon alienates readers. Nature 2020;579(7798):309. Crossref, Medline, Google Scholar3. Anvari A, Halpern EF, Samir AE. Statistics 101 for Radiologists. RadioGraphics 2015;35(6):1789–1801. Link, Google Scholar4. Scientific Style Guide: Writing a Manuscript for Radiology. Radiological Society of North America Web site.https://pubs.rsna.org/page/radiology/author-instructions/scientificediting. Accessed June 7, 2022. Google ScholarArticle HistoryPublished online: Aug 09 2022Published in print: Dec 2022 FiguresReferencesRelatedDetailsCited ByIntroduction to Research for International Young Academics: A Life-changing Experience Beyond All ExpectationsAlessia Guarnera, 6 June 2023 | Radiology, Vol. 307, No. 5Importance of Reducing Barriers and Promoting Frontier-Free ResearchJosé David Cardona Ortegón, Sergio Valencia, Javier Ricardo Ortiz Llinás, Karen Cifuentes Gaitán, 28 March 2023 | Radiology, Vol. 307, No. 4Looking Back at 2022 and ahead to 2023 for the Korean Journal of RadiologySeong HoPark2023 | Korean Journal of Radiology, Vol. 24, No. 1Recommended Articles Low-Dose CT Screening for Lung Cancer: Evidence from 2 Decades of StudyRadiology: Imaging Cancer2020Volume: 2Issue: 2An Update on Coronary Artery Calcium Interpretation at Chest and Cardiac CTRadiology: Cardiothoracic Imaging2021Volume: 3Issue: 1Mediastinal Lymphadenopathy in the National Lung Screening Trial (NLST) Is Associated with Interval Lung CancerRadiology2021Volume: 302Issue: 3pp. 684-692Early-Stage Non–Small Cell Lung Cancer: Quantitative Imaging Characteristics of 18F Fluorodeoxyglucose PET/CT Allow Prediction of Distant MetastasisRadiology2016Volume: 281Issue: 1pp. 270-278Incidental Lymphadenopathy at CT Lung Cancer ScreeningRadiology2021Volume: 302Issue: 3pp. 693-694See More RSNA Education Exhibits “Why-RADS”, A 2021 Update On The Standardized Reporting And Data SystemsDigital Posters202112 Angry Men: A Comprehensive Case-based Radiologic Review of Twelve Smoking-Related Cancers and BeyondDigital Posters2019Image-Guided Lung Biopsy in the Era of Personalized Cancer Care: What Interventional Radiologists Need to KnowDigital Posters2020 RSNA Case Collection Small cell lung carcinomaRSNA Case Collection2020Anomalous origin of the right coronary artery from the left sinus of ValsalvaRSNA Case Collection2020Hypertrophic OsteoarthropathyRSNA Case Collection2020 Vol. 305, No. 3 Metrics Altmetric Score PDF download