Empirical assessment of effect of publication bias on meta-analysesBMJ 2000; 320 doi: https://doi.org/10.1136/bmj.320.7249.1574 (Published 10 June 2000) Cite this as: BMJ 2000;320:1574
- A J Sutton, Lecturer in medical statistics ()a,
- S J Duval, assistant professorb,
- R L Tweedie, headc,
- K R Abrams, senior lecturer in medical statisticsa,
- D R Jones, professor in medical statisticsa
- a Department of Epidemiology and Public Health, University of Leicester, Leicester LE1 6TP
- b Division of Epidemiology, School of Public Health, University of Minnesota, Minneapolis, USA
- c Department of Biostatistics, School of Public Health, University of Minnesota
- Correspondence to: A J Sutton
- Accepted 2 March 2000
Objective: To assess the effect of publication bias on the results and conclusions of systematic reviews and meta-analyses.
Design: Analysis of published meta-analyses by trim and fill method.
Studies: 48 reviews in Cochrane Database of Systematic Reviews that considered a binary endpoint and contained 10 or more individual studies.
Main outcome measures: Number of reviews with missing studies and effect on conclusions of meta-analyses.
Results: The trim and fill fixed effects analysis method estimated that 26 (54%) of reviews had missing studies and in 10 the number missing was significant. The corresponding figures with a random effects model were 23 (48%) and eight. In four cases, statistical inferences regarding the effect of the intervention were changed after the overall estimate for publication bias was adjusted for.
Conclusions: Publication or related biases were common within the sample of meta-analyses assessed. In most cases these biases did not affect the conclusions. Nevertheless, researchers should check routinely whether conclusions of systematic reviews are robust to possible non-random selection mechanisms.
Funding None declared.
Competing interests None.
Figures illustrating the method and funnel plots of all trials and a full table of results are available on the BMJ's website