Cited by 1,133Meta-analysis and systematic reviewsStatistical Methods in Clinical TrialsPharmaceutical industry and healthcare

The impact of outcome reporting bias in randomised controlled trials on a cohort of systematic reviews

Open access
Kirkham, J. J, Dwan, K. M, Altman, D. G, Gamble, C., Dodd, S., Smyth, R., Williamson, P. R·2010·BMJ

Abstract

OBJECTIVE: To examine the prevalence of outcome reporting bias-the selection for publication of a subset of the original recorded outcome variables on the basis of the results-and its impact on Cochrane reviews. DESIGN: A nine point classification system for missing outcome data in randomised trials was developed and applied to the trials assessed in a large, unselected cohort of Cochrane systematic reviews. Researchers who conducted the trials were contacted and the reason sought for the non-reporting of data. A sensitivity analysis was undertaken to assess the impact of outcome reporting bias on reviews that included a single meta-analysis of the review primary outcome. RESULTS: More than half (157/283 (55%)) the reviews did not include full data for the review primary outcome of interest from all eligible trials. The median amount of review outcome data missing for any reason was 10%, whereas 50% or more of the potential data were missing in 70 (25%) reviews. It was clear from the publications for 155 (6%) of the 2486 assessable trials that the researchers had measured and analysed the review primary outcome but did not report or only partially reported the results. For reports that did not mention the review primary outcome, our classification regarding the presence of outcome reporting bias was shown to have a sensitivity of 88% (95% CI 65% to 100%) and specificity of 80% (95% CI 69% to 90%) on the basis of responses from 62 trialists. A third of Cochrane reviews (96/283 (34%)) contained at least one trial with high suspicion of outcome reporting bias for the review primary outcome. In a sensitivity analysis undertaken for 81 reviews with a single meta-analysis of the primary outcome of interest, the treatment effect estimate was reduced by 20% or more in 19 (23%). Of the 42 meta-analyses with a statistically significant result only, eight (19%) became non-significant after adjustment for outcome reporting bias and 11 (26%) would have overestimated the treatment effect by 20% or more. CONCLUSIONS: Outcome reporting bias is an under-recognised problem that affects the conclusions in a substantial proportion of Cochrane reviews. Individuals conducting systematic reviews need to address explicitly the issue of missing outcome data for their review to be considered a reliable source of evidence. Extra care is required during data extraction, reviewers should identify when a trial reports that an outcome was measured but no results were reported or events observed, and contact with trialists should be encouraged.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1136/bmj.c365

Cites

23 of 23 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Contacting authors or manufacturers may be necessary when publications, conference proceedings, or clinical trials registry records do not provide the complete information needed.

    “Contacting authors or manufacturers may also be necessary when publications, conference proceedings, or clinical trials registry records do not provide the complete information needed [63, 81].”

    From PRISMA-S: an extension to the PRISMA Statement for Reporting Literature Searches in Systematic Reviews (2021) · cites this paper as [81]

    In “Implications for systematic reviews” (page not provided), the paper advises contacting trialists to obtain outcome results missing from trial reports. It does not address contacting manufacturers or specifically recommend contact when conference proceedings or registry records are incomplete. (LLM, conf=0.85)

    Experimental — accuracy not yet measured
  • Funnel plot asymmetry may be explained by true heterogeneity of effect with respect to study size, poor methodological design (and hence bias of small studies), and selective non-reporting.

    “Several explanations may account for funnel plot asymmetry, including true heterogeneity of effect with respect to study size, poor methodological design (and hence bias of small studies), and selective non‐reporting (Kirkham 2010).”

    From Mindfulness-based interventions for adults with type 2 diabetes mellitus (2021) · cites this paper as Kirkham 2010

    In “Accuracy of classification” (page unavailable), the paper describes a method that assumes smaller studies are more likely than larger studies to omit outcome results, supporting a link between study size and selective non-reporting. The passages do not discuss funnel plot asymmetry or identify true effect heterogeneity and poor methodological design as explanations for it. (LLM, conf=0.76)

    Experimental — accuracy not yet measured
  • Where outcome misreporting is permitted, it increases the likelihood that reported differences have arisen through chance or are exaggerated.

    “Where outcome misreporting is permitted, it increases the likelihood that reported differences have arisen through chance or are exaggerated [2, 3].”
  • Selective reporting of outcomes (and predictors) is often a risk.

    “However, selective reporting of outcomes (and predictors) is often a risk [51].”
  • Outcome reporting bias was assessed by integrating results of Appendix 6 and Appendix 7.

    “We assessed outcome reporting bias by integrating the results of Appendix 6, 'Matrix of trial endpoints (publications and trial documents)' (Mathieu 2009), with those of Appendix 7 'High risk of outcome reporting bias according to ORBIT classification' (Kirkham 2010).”

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Authors
Kirkham, J. J., Dwan, K. M., Altman, D. G., Gamble, C., Dodd, S., Smyth, R., & Williamson, P. R.
Affiliations
  • University of Liverpool (GB)
  • University of Oxford (GB)
Corresponding author
Williamson, P. R
Venue
BMJ
Publisher
BMJ
Published
2010-02-15
Volume
340
Issue
feb15 1
Pages
c365-c365
ISSN
  • 0007-1447 (ISSN-L)
  • 0959-8138
  • 1468-5833
OpenAlex
W2046391772
Cited by
1,133
References
23
Citations by year
2026: 27 · 2025: 37 · 2024: 53 · 2023: 46 · 2022: 58 · 2021: 77 · 2020: 76 · 2019: 72 · 2018: 66 · 2017: 95 · 2016: 102 · 2015: 111 · 2014: 97 · 2013: 84 · 2012: 70
FWCI
18.91
Type
journal article
Language
en
Open access
Locations
  • BMJ (publishedVersion)
  • PubMed (publishedVersion)
  • Research Explorer (The University of Manchester) (publishedVersion)
  • HighWire Press Open Archive (submittedVersion)
Topics
  • Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › Meta-analysis and systematic reviews
  • Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods in Clinical Trials
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Pharmacology › Pharmaceutical industry and healthcare
Keywords
selective reporting, missing outcome data, publication bias, randomized controlled trials, meta-analysis, sensitivity analysis, data extraction, treatment effect estimation
MeSH terms
Data Collection, Outcome and Process Assessment, Health Care, Research Design, Review Literature as Topic, Cohort Studies, Bias, Randomized Controlled Trials as Topic

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review