RCT indicates randomized clinical trial.
eAppendix. MEDLINE detailed search strategy
Customize your JAMA Network experience by selecting one or more topics from the list below.
Identify all potential conflicts of interest that might be relevant to your comment.
Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.
Err on the side of full disclosure.
If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.
Not all submitted comments are published. Please see our commenting policy for details.
Khan MS, Lateef N, Siddiqi TJ, et al. Level and Prevalence of Spin in Published Cardiovascular Randomized Clinical Trial Reports With Statistically Nonsignificant Primary Outcomes: A Systematic Review. JAMA Netw Open. Published online May 03, 20192(5):e192622. doi:10.1001/jamanetworkopen.2019.2622
Do authors of cardiovascular randomized clinical trials present statistically nonsignificant primary outcomes accurately and objectively?
In this systematic review that included 93 reports of randomized clinical trials from 6 high-impact journals, positive spin of statistically nonsignificant primary outcomes was found in 57% of abstracts and 67% of main text of the published articles.
Despite peer review, manipulation of language in the cardiovascular literature is common and may have implications for scientific integrity, patient care, peer review, and medical progress.
Clinical researchers are obligated to present results objectively and accurately to ensure readers are not misled. In studies in which primary end points are not statistically significant, placing a spin, defined as the manipulation of language to potentially mislead readers from the likely truth of the results, can distract the reader and lead to misinterpretation and misapplication of the findings.
To determine the level and prevalence of spin in published reports of cardiovascular randomized clinical trial (RCT) reports.
MEDLINE was searched from January 1, 2015, to December 31, 2017, using the Cochrane highly sensitive search strategy.
Inclusion criteria were parallel-group RCTs published from January 1, 2015, to December 31, 2017 in 1 of 6 high-impact journals (New England Journal of Medicine, The Lancet, JAMA, European Heart Journal, Circulation, and Journal of the American College of Cardiology) with primary outcomes that were not statistically significant were included in the analysis.
Data Extraction and Synthesis
Analysis began in August 2018. Data were extracted and verified by 2 independent investigators using a standard collection form. In cases of disagreement between the 2 investigators, a third investigators served as arbitrator.
Main Outcomes and Measures
The classifications of spin type, severity, and extent were determined according to predefined criteria. Primary clinical outcomes were divided into safety of treatment, efficacy of treatment, and both.
Of 587 studies identified, 93 RCT reports (15.8%) met inclusion criteria. Spin was identified in 53 abstracts (57%; 95% CI, 47%-67%) and 62 main texts of published articles (67%; 95% CI, 57%-75%). Ten reports (11%; 95% CI, 6%-19%) had spin in the title, 35 reports (38%; 95% CI, 28%-48%) had spin in the results section, and 50 reports (54%; 95% CI, 44%-64%) had spin in the conclusions. Among the abstracts, spin was observed in 38 results sections (41%; 95% CI, 31%-51%) and 45 conclusions sections (48%; 95% CI, 38%-58%).
Conclusions and Relevance
This study suggests that in reports of cardiovascular RCTs with statistically nonsignificant primary outcomes, investigators often manipulate the language of the report to detract from the neutral primary outcomes. To best apply evidence to patient care, consumers of cardiovascular research should be aware that peer review does not always preclude the use of misleading language in scientific articles.
Randomized clinical trials (RCTs) are considered the criterion standard for determining efficacy of an intervention. Methodologically rigorous RCTs generate the highest certainty of evidence and form the basis of clinical guidelines.1 Evidence-based practice depends on accurate presentation of a trial’s results. Journal editors allow authors of scientific articles broad latitude in the use of language when reporting their study, which may subconsciously or consciously shape the impression of their results for readers.
Spin is defined as the manipulation of language to potentially mislead readers from the likely truth of the results.2-4 Boutron et al5 developed a technique to identify and classify spin in RCT reports. Their approach focused on RCTs reporting statistically nonsignificant primary outcomes because the interpretation of these results is more likely to be subject to prior beliefs of effectiveness, leading to potential bias in reporting.6 Subsequent studies have used similar approaches to systematically assess the explicit presentation of nonsignificant results in trial reports in various subspecialties.7,8 Although previous articles have briefly highlighted distorted presentation and interpretation of results in the cardiovascular (CV) literature,9,10 to our knowledge, these strategies have not been systematically identified and evaluated. Such reporting bias may lead to inaccurate estimation of the benefit of CV interventions and affect patient care. We aimed to assess the prevalence and extent of spin in a cohort of CV RCT reports.
This systematic review is reported following the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline and the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies.11,12
We searched MEDLINE using the Cochrane highly sensitive search strategy13 for CV RCTs published from January 1, 2015, to December 31, 2017, in 6 high-impact journals (New England Journal of Medicine, The Lancet, JAMA, European Heart Journal, Circulation, and Journal of the American College of Cardiology). A detailed search strategy is provided in the Appendix in the Supplement. We included only RCTs with parallel groups (defined as studies in which each participant is randomized to 1 of the intervention arms) and clearly identified primary outcomes (ie, distinctly mentioned in the trial as the primary study objective) that were not statistically significant (ie, P ≥ .05).
We excluded pilot studies, brief communications, research letters, factorial and split-body designs, cluster trials, equivalence or noninferiority trials, crossover trials, multigroup trials, and phase 1 or 2 trials. Trials that exclusively focused on economic evaluations and diagnostic test accuracy were also excluded. The definitions and strategies for the genre of trials reflect those used by Chan and Altman.14
The preliminary search identified 2473 RCT reports cited in PubMed, which were then transferred to EndNote (Clarivate Analytics). The titles and abstracts of the identified studies were then screened to exclude irrelevant studies. Full-text studies were subsequently obtained and evaluated for the remaining 1166 reports. After assessing for relevance, 587 studies were included. The screening process was carried out by 2 independent reviewers (N.L. and T.J.S.), and a third reviewer (M.S.K.) was consulted in the event of discrepancies.
For each selected RCT report, we extracted 2016 journal impact factor, number of citations in PubMed until August 2018, source of funding, intervention in the active and control groups, primary clinical outcome, and presence or absence of conflict of interest of the first and last authors. Presence of conflict of interest for the first and last authors was categorized as yes or no. Two reviewers (N.L. and T.J.S.) extracted all relevant data onto a standardized data collection form and then independently read the title, abstract and methods, results, discussion, and conclusions sections to identify the type, severity, and extent of spin, if any. Any disagreements were resolved through discussion. When a consensus could not be reached, a third author (M.S.K.) arbitrated. The classifications of spin type, severity, and extent were based on the article by Boutron et al.5 All the authors were briefed about these criteria before data collection and all unanimously agreed to its application.
The κ coefficient is a measure of the extent of agreement between the 2 independent investigators and was calculated using the methodology outlined by Landis and Koch.15 The frequency of disagreements and agreements between the reviewers were computed in the Kappa Calculator (Statistics Solutions),16 and the κ statistics were determined for each outcome. There was substantial agreement in reproducibility for presence of spin, with κ = 0.77 (95% CI, 0.52-0.91) in the main text conclusions section, κ = 0.73 (95% CI, 0.41-0.87) in the abstract conclusions section, κ = 0.79 (95% CI, 0.53-0.90) in the main text results section, κ = 0.64 (95% CI, 0.42-0.79) in the abstract results section, κ = 0.75 (95% CI, 0.57-0.96) in the discussion section, and κ = 0.65 (95% CI, 0.57-0.89) for presence of spin in the title.
Primary clinical outcomes were divided into 3 categories: safety of treatment, efficacy of treatment, and both. Randomized clinical trials in which the primary outcome was related to measurement of adverse events from a treatment were classified under safety of treatment. Trials in which the primary outcome was related to measure of effectiveness of a drug (eg, reduction in mortality or glycemic or blood pressure levels) were classified under efficacy of treatment. We also divided the end point by type: dichotomous, other, or both.
Following the methods outlined by Boutron et al,5 we defined spin as the “use of specific reporting strategies, from whatever motive, to highlight that the experimental treatment is beneficial, despite a statistically nonsignificant difference for the primary outcome [ie, inappropriate use of causal language], or to distract the reader from statistically nonsignificant results [ie, to focus on a statistically significant secondary result].”5 Using a prespecified spin classification scheme, the following sections of each trial were scrutinized to determine whether authors had used spin: the title, the results and conclusions sections of the abstract, and the results, discussion, and conclusions sections of the main text.6 We then identified which of 3 specific spin strategies had been used: (1) authors pivoted on statistically significant secondary results in the form of focus on within-group comparison, secondary outcomes, subgroup, or per-protocol analyses; (2) authors interpreted statistically nonsignificant results of the primary outcomes to show treatment equivalence or to rule out an adverse event; and (3) authors emphasized the beneficial effect of the treatment with or without acknowledging the statistically nonsignificant primary outcome. Strategies of spin that could not be classified under 1 of the 3 schemes were systematically recorded as other.
We also assessed the level of spin in the abstract and main text conclusions section. We classified the level of spin as high, moderate, or low, with none acting as a default category. High spin was defined as no acknowledgment of the statistically nonsignificant primary outcome, no uncertainty in the framing, and no recommendations for further trials. Moderate spin was defined as no acknowledgment of the nonsignificant primary outcome but some uncertainty in the framing or the presence of recommendations for further trials. Low spin was defined as acknowledgment of the nonsignificant primary outcome or uncertainty in the framing and presence of recommendations for further trials. The purpose of calculating level of spin was to assess the heterogeneity of reporting of spin in the conclusions section.
We calculated the extent of spin across the whole of each report, defined as number of sections with spin in both the abstract and main text. Spin in the title was excluded from this calculation. We used 4 categories to classify the extent of spin: (1) spin in 1 section other than the conclusions section; (2) spin in the conclusions section only; (3) spin in 2 sections, not including the abstract; and (4) spin in all sections.
We calculated the number and percentage of reports with 95% CIs for categorical variables and medians with interquartile ranges for continuous variables. The χ2 linear-by-linear association test was conducted to assess the association of level of spin in the abstract or text with first or last author conflict of interest. The Spearman ρ correlation test was performed to evaluate any correlation between spin in the text or abstract and the citations per year of each report. Finally, the Kruskal-Wallis test with Dunn post hoc method was used to evaluate whether spin was associated more with a certain type of end point, type of primary outcome, type of experimental treatment (eg, drug, surgical operation or procedure, or device), comparator (eg, placebo, drug, or device), subspecialty, funding source, or journal. All statistical analyses were performed using SPSS Statistics version 23.0 (IBM). P values were 2-tailed, and a P value less than .05 was considered statistically significant.
Of 587 studies included in the sample, we identified 229 parallel-group RCTs with clearly identified primary outcomes. Of these, 93 studies reported statistically nonsignificant results for their primary outcome (Figure). We identified 25 studies (27%; 95% CI, 19%-37%) funded solely by for-profit sources and 37 studies (40%; 95% CI, 30%-50%) funded solely by nonprofit sources. Detailed characteristics of the included RCTs are outlined in Table 1.
Table 2 outlines the frequency of spin observed in each section of the abstracts and texts. Spin was observed in the titles of 10 reports (11%; 95% CI, 6%-19%). In the abstracts, we identified spin in 38 results sections (41%; 95% CI, 31%-51%) and in 45 conclusions sections (48%; 95% CI, 38%-58%). In the main texts, 35 results sections (38%; 95% CI, 28%-48%) and 35 discussion sections (38%; 95% CI, 28%-48%) showed spin. Moreover, 50 of the main text conclusions sections (54%; 95% CI, 44%-64%) included spin.
The strategies of spin in each section of the abstract and full text are presented in detail in Table 2. In the abstracts, spin in the results sections was found in 13 reports (14%; 95% CI, 8%-22%) focusing on statistically significant secondary outcomes and in 10 reports (11%; 95% CI, 6%-19%) focusing on statistically significant subgroup analyses, whereas 17 reports (18%; 95% CI, 12%-27%) that used spin in the conclusions section mainly focused on reporting treatment effectiveness. In the main texts, spin in the results section included 11 reports (12%; 95% CI, 7%-20%) presenting statistically significant secondary outcomes and in the conclusions section in 16 reports (17%; 95% CI, 11%-26%) focusing on treatment effectiveness.
The extent of spin varied among our cohort (Table 3). Overall, 53 abstracts (57%; 95% CI, 47%-67%) and 62 main texts (67%; 95% CI, 57%-75%) had spin in at least 1 section. Moreover, 26 abstracts (28%; 95% CI, 20%-38%) and 18 main texts (19%; 95% CI, 12%-29%) had spin in all of their sections.
Table 3 outlines the level of spin in the conclusions sections of the abstract and main text. About half the conclusions sections included spin, with 41 abstracts (44%; 95% CI, 34%-54%) and 49 full texts (53%; 95% CI, 43%-62%) containing some level of spin. In reports where spin was identified in the conclusions, it was mostly low-level spin.
Conflicts of interest disclosures of the first authors (abstract, χ2 = 0.215, P = .64; full text, χ2 = 0.003, P = .96) and last authors (abstract, χ2 = 1.675, P = .20; full text, χ2 = 2.644, P = .10) did not correlate with spin. A modest but significant negative correlation was present between citations per year and the level of spin in the abstract (ρ = −0.20; P = .03) and main text (ρ = −0.3; P = .049).
In the main text of the articles, we found that spin was not significantly associated with CV subspecialty, industry funding, or type of journal. Articles in some journals were found to have more spin than those in other journals. However, given the relatively small number of articles from each journal, this finding should be considered exploratory and must be viewed with caution. Table 4 details these associations.
In RCTs with statistically nonsignificant primary outcomes published in high-impact CV journals, we found considerable manipulation of language in both the abstracts and the full texts of the RCT reports. This occurred in RCTs that received industry or public funding. Our results align well with findings of spin in reports from other medical areas.6,10
When we found spin in the results and discussion sections of studies with nonsignificant primary efficacy outcomes, authors tended to focus on statistically significant secondary end points, within-group analyses, and subgroup analyses. In some cases of spin, within-group comparisons showed statistical significance in the experimental arm but not in the comparator, which was interpreted as a potential benefit of the treatment. Other ways authors used spin were to report lack of harm from the safety data without highlighting the statistically nonsignificant efficacy result, to focus on effectiveness of both treatment arms when a statistically significant change was seen from baseline for each group, or to use both these methods to spin statistically nonsignificant results.
Unlike previous studies on spin, we did not observe an association of the level of spin with conflicts of interest disclosures from the first and last authors.17,18 More specifically, industry-funded research had a lower proportion of spin than nonprofit-funded research. This is important because it goes against a commonly expressed view that industry funding may have a direct or indirect effect on an investigator to explicitly describe nonsignificant results.19-22 However, a 2009 report23 highlighted high rates of underreporting of financial conflicts of interest by investigators. It is possible that researchers who are prone to spin may intentionally or unintentionally underreport their financial relationships.
We can only speculate on the reason authors use positive spin. Incentives likely play a role. Publication in high-impact journals fosters career advancement and future grant funding. A 2014 study24 on publication bias has shown that positive findings are more likely to be published in higher-impact journals. Authors may naturally tend to accentuate a positive approach to their results. We believe spin may decrease if neutral results were as likely to be published.
In some circumstances, additional spin beyond what is actually in the report may occur when it is published. For example, an accompanying press release by the journal, the authors’ institution(s), or the funding agency may focus on any positive aspects without mentioning the primary outcome. Furthermore, in some instances the funding agency may employ media coaches to teach investigators how to stay on message.
These observations have significant implications for the integrity of clinical science, the translation of clinical evidence at the bedside, peer review, and the rate of medical progress. Manipulation of language to distort findings may also lead to further public distrust in science. For example, it is estimated that around 5% to 10% of individuals have strong antivaccination beliefs.25 Among these individuals, lack of trust in health professionals or in public health institutions can be pivotal in making decisions about vaccinating their infants.26
Our findings support efforts to improve higher-quality reporting of scientific studies.27 The high prevalence of manipulation of language in the literature has great importance for consumers of medical evidence, which, given the freedom of information in the digital age, affects both patients as well as clinicians. Readers of medical evidence should be aware of these rhetorical techniques in science publications.28-30 More robust editorial and peer review may help reduce the level of spin. Finally, we propose that medical progress might move faster and waste fewer resources if authors were able to publish neutral findings in higher-impact journals.31 Knowing what does not work is also an important part of scientific discovery.
Our review has limitations that need to be considered when interpreting the results. First, although our study cohort is a representative sample of high-impact RCT reports, our search was limited to a short period in select journals. Second, we only selected studies with clearly defined primary outcomes and excluded studies without a clearly specified primary outcome, as the latter would not allow us to identify spin. Third, we were not able to assess from the data available which spin strategies were more successful, and neither were we able to conclude why spin occurs in medical research. Utmost efforts were made to avoid bias by following strict criteria for identification and assessment of spin by 2 independent reviewers. However, the presence of subjective bias cannot be completely excluded. Furthermore, we were not able to evaluate the effect of spin on peer reviewers, editors, and health care professionals on the receiving end. While this study is unique in the CV literature, to our knowledge, several unanswered questions arise from our findings, paving the way for future research to examine the mechanisms of spin and implications of trial reporting in the field of CV medicine.
Our findings show that in approximately 67% of CV RCT reports, the reporting and interpretation of outcomes is inconsistent with actual results in at least 1 section of the article. Consumers of CV research should become familiar with the principles of evidence-based medicine to appraise and appropriately apply trial evidence.
Accepted for Publication: March 1, 2019.
Published: May 3, 2019. doi:10.1001/jamanetworkopen.2019.2622
Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2019 Khan MS et al. JAMA Network Open.
Corresponding Author: Richard A. Krasuski, MD, Department of Cardiovascular Medicine, Duke University Health System, DUMC 3012, 220 Erwin Rd, Durham, NC 27710 (firstname.lastname@example.org).
Author Contributions: Drs M. S. Khan and Lateef had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: M. S. Khan, Lateef, Rehman, Alnaimat, Riaz, Murad, Mandrola, Krasuski.
Acquisition, analysis, or interpretation of data: M. S. Khan, Siddiqi, Rehman, Alnaimat, S. U. Khan, Doukky.
Drafting of the manuscript: M. S. Khan, Lateef, Siddiqi, Rehman, Alnaimat, Murad, Krasuski.
Critical revision of the manuscript for important intellectual content: M. S. Khan, Siddiqi, Alnaimat, S. U. Khan, Riaz, Murad, Mandrola, Doukky, Krasuski.
Statistical analysis: M. S. Khan, Siddiqi, Rehman.
Administrative, technical, or material support: Alnaimat, Mandrola, Krasuski.
Supervision: M. S. Khan, Alnaimat, Riaz, Murad, Doukky, Krasuski.
Conflict of Interest Disclosures: Dr Krasuski reported receiving grants from Edwards Lifesciences and Abbott; receiving grants and personal fees from Actelion and serving as a nonfunded scientific advisor for from Ventripoint outside the submitted work. No other disclosures were reported.
Create a personal account or sign in to: