Disclosure of data and statistical commands should accompany completely reported studies

Automated detection of over- and under-dispersion in baseline tables in randomised controlled trials

Background: Papers describing the results of a randomised trial should include a baseline table that compares the characteristics of randomised groups. Researchers who fraudulently generate trials often unwittingly create baseline tables that are implausibly similar (under-dispersed) or have large differences between groups (over-dispersed). I aimed to create an automated algorithm to screen for under- and over-dispersion in the baseline tables of randomised trials. Methods: Using a cross-sectional study I examined 2,245 randomised controlled trials published in health and medical journals on PubMed Central. I estimated the probability that a trial's baseline summary statistics were under- or over-dispersed using a Bayesian model that examined the distribution of t-statistics for the between-group differences, and compared this with an expected distribution without dispersion. I used a simulation study to test the ability of the model to find under- or over-dispersion and compared its performance with an existing test of dispersion based on a uniform test of p-values. My model combined categorical and continuous summary statistics, whereas the uniform test used only continuous statistics. Results: The algorithm had a relatively good accuracy for extracting the data from baseline tables, matching well on the size of the tables and sample size. Using t-statistics in the Bayesian model out-performed the uniform test of p-values, which had many false positives for skewed, categorical and rounded data that were not under- or over-dispersed. For trials published on PubMed Central, some tables appeared under- or over-dispersed because they had an atypical presentation or had reporting errors. Some trials flagged as under-dispersed had groups with strikingly similar summary statistics. Conclusions: Automated screening for fraud of all submitted trials is challenging due to the widely varying presentation of baseline tables. The Bayesian model could be useful in targeted checks of suspected trials or authors.

Automated detection of over- and under-dispersion in baseline tables in randomised controlled trials

Background: Papers describing the results of a randomised trial should include a baseline table that compares the characteristics of randomised groups. Researchers who fraudulently generate trials often unwittingly create baseline tables that are implausibly similar (under-dispersed) or have large differences between groups (over-dispersed). I aimed to create an automated algorithm to screen for under- and over-dispersion in the baseline tables of randomised trials. Methods: Using a cross-sectional study I examined 2,245 randomised controlled trials published in health and medical journals on PubMed Central. I estimated the probability that a trial's baseline summary statistics were under- or over-dispersed using a Bayesian model that examined the distribution of t-statistics for the between-group differences, and compared this with an expected distribution without dispersion. I used a simulation study to test the ability of the model to find under- or over-dispersion and compared its performance with an existing test of dispersion based on a uniform test of p-values. My model combined categorical and continuous summary statistics, whereas the uniform test used only continuous statistics. Results: The algorithm had a relatively good accuracy for extracting the data from baseline tables, matching well on the size of the tables and sample size. Using t-statistics in the Bayesian model out-performed the uniform test of p-values, which had many false positives for skewed, categorical and rounded data that were not under- or over-dispersed. For trials published on PubMed Central, some tables appeared under- or over-dispersed because they had an atypical presentation or had reporting errors. Some trials flagged as under-dispersed had groups with strikingly similar summary statistics. Conclusions: Automated screening for fraud of all submitted trials is challenging due to the widely varying presentation of baseline tables. The Bayesian model could be useful in targeted checks of suspected trials or authors.

Authors of trials from high-ranking anesthesiology journals were not willing to share raw data

OBJECTIVES

To analyze data sharing practices among authors of randomized controlled trials (RCTs) published in seven high-ranking anesthesiology journals from 2014 to 2016.

STUDY DESIGN AND SETTING

We analyzed data sharing statements in 619 included RCTs and contacted their corresponding authors, asking them to share de-identified raw data from trial.

RESULTS

Of the 86 (14%) authors who responded to our query for data sharing, only 24 (4%) provided the requested data. Only one of those 24 had a data sharing statement in the published manuscript. Only 24 (4%) of manuscripts contained statements suggesting a willingness to share trial data; only one of those authors actually shared data. There was no difference in proportion of data sharing between studies with commercial and nonprofit funding. Among the 62 authors who refused to provide data, reasons were seldom provided. When reasons were provided, common themes included issues regarding data ownership and participant privacy. Only one of the seven analyzed journals encouraged authors toward data sharing.

CONCLUSION

Willingness to share data among anesthesiology RCTs is very low. To achieve widespread availability of de-identified trial data, journals should request their publication, as opposed to only encouraging authors to do so.

Reproducible research practices are underused in systematic reviews of biomedical interventions

OBJECTIVES

To evaluate how often reproducible research practices, which allow others to recreate the findings of studies, given the original data, are used in systematic reviews (SRs) of biomedical research.

STUDY DESIGN AND SETTING

We evaluated a random sample of SRs indexed in MEDLINE during February 2014, which focused on a therapeutic intervention and reported at least one meta-analysis. Data on reproducible research practices in each SR were extracted using a 26-item form by one author, with a 20% random sample extracted in duplicate. We explored whether the use of reproducible research practices was associated with an SR being a Cochrane review, as well as with the reported use of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement.

RESULTS

We evaluated 110 SRs of therapeutic interventions, 78 (71%) of which were non-Cochrane SRs. Across the SRs, there were 2,139 meta-analytic effects (including subgroup meta-analytic effects and sensitivity analyses), 1,551 (73%) of which were reported in sufficient detail to recreate them. Systematic reviewers reported the data needed to recreate all meta-analytic effects in 72 (65%) SRs only. This percentage was higher in Cochrane than in non-Cochrane SRs (30/32 [94%] vs. 42/78 [54%]; risk ratio 1.74, 95% confidence interval 1.39-2.18). Systematic reviewers who reported imputing, algebraically manipulating, or obtaining some data from the study author/sponsor infrequently stated which specific data were handled in this way. Only 33 (30%) SRs mentioned access to data sets and statistical code used to perform analyses.

CONCLUSION

Reproducible research practices are underused in SRs of biomedical interventions. Adoption of such practices facilitates identification of errors and allows the SR data to be reanalyzed.

Reporting guidance considerations from a statistical perspective: overview of tools to enhance the rigour of reporting of randomised trials and systematic reviews

OBJECTIVE

Research waste has received considerable attention from the biomedical community. One noteworthy contributor is incomplete reporting in research publications. When detailing statistical methods and results, ensuring analytic methods and findings are completely documented improves transparency. For publications describing randomised trials and systematic reviews, guidelines have been developed to facilitate complete reporting. This overview summarises aspects of statistical reporting in trials and systematic reviews of health interventions.

METHODS

A narrative approach to summarise features regarding statistical methods and findings from reporting guidelines for trials and reviews was taken. We aim to enhance familiarity of statistical details that should be reported in biomedical research among statisticians and their collaborators.

RESULTS

We summarise statistical reporting considerations for trials and systematic reviews from guidance documents including the Consolidated Standards of Reporting Trials (CONSORT) Statement for reporting of trials, the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) Statement for trial protocols, the Statistical Analyses and Methods in the Published Literature (SAMPL) Guidelines for statistical reporting principles, the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement for systematic reviews and PRISMA for Protocols (PRISMA-P). Considerations regarding sharing of study data and statistical code are also addressed.

CONCLUSIONS

Reporting guidelines provide researchers with minimum criteria for reporting. If followed, they can enhance research transparency and contribute improve quality of biomedical publications. Authors should employ these tools for planning and reporting of their research.

Prevention of selective outcome reporting: let us start from the beginning

BACKGROUND

Healthcare professionals and patients could be negatively influenced in their judgments by articles and meta-analyses presenting selective outcome reporting. Clinical trials should be transparent from inception to the publication of results. To this end, trial prospective registration is an ethical and scientific requirement that have shown to be effective in preventing selective reporting of outcomes. However, even journals with a clear pre-registration policy publish trial results that were retrospectively registered.

SITUATION

Analyses of registration of randomized clinical trials recently published in top specialty journals and of meta-analyses with suspicion of including trials with outcome reporting bias have shown that retrospective registration is in the range from 56 to 76 %. This translates into publication of primary endpoints that differ from those included in the registry: some 30 % of trials showed discrepancies between the primary endpoint in the trial registry and the article. Furthermore, it has been shown that 8 % of all clinical trials published by 6 high-impact ICMJE-member journals was retrospectively registered after primary endpoint ascertainment could have had taken place, raising concerns that endpoints may not have been pre-specified, or were changed. With regards to meta-analyses, 34 % of Cochrane systematic reviews included one or more trials with a high suspicion of selective reporting bias for the primary outcome.

PROPOSAL

Retrospective registration of trials may foster selective outcome reporting unless journal editors implement specific quality control processes aiming to prevent or minimize this type of bias. Prospective registration of trials-and protocol public disclosure if proven effective in future studies-prevents outcome reporting bias, a must to ensure clinicians and patients have access to reliable clinical trial results. Journal editors should enforce, rather than encourage, appropriate measures to ensure publication of trials free of outcome reporting bias.