This is an archived version. For the current version, please go to training.cochrane.org/handbook/current.

Although there is clear evidence that publication and other reporting biases lead to over-optimistic estimates of intervention effects, overcoming, detecting and correcting for publication bias is problematic. Comprehensive searches are important, particularly to identify studies as well defined as randomized trials. However, comprehensive searching is not sufficient to prevent some substantial potential biases.

Publication bias should be seen as one of a number of possible causes of ‘small-study effects’ – a tendency for estimates of the intervention effect to be more beneficial in smaller studies. Funnel plots allow review authors to make a visual assessment of whether small-study effects may be present in a meta-analysis. For continuous (numerical) outcomes with intervention effects measured as mean differences, funnel plots and statistical tests for funnel plot asymmetry are valid. However for dichotomous outcomes with intervention effects expressed as odds ratios, the standard error of the log odds ratio is mathematically linked to the size of the odds ratio, even in the absence of small-study effects. This can cause funnel plots plotted using log odds ratios (or odds ratios on a log scale) to appear asymmetric and can mean that P values from the test of Egger et al. are too small. For other effect measures, firm guidance is not yet offered. Three statistical tests for small-study effects are recommended for use in Cochrane reviews, provided that there are at least 10 studies. However, none is implemented in RevMan and statistical support is usually required. Only one test has been shown to work when the between-study heterogeneity variance exceeds 0.1. Results from tests for funnel plot asymmetry should be interpreted cautiously. When there is evidence of small-study effects, publication bias should be considered as only one of a number of possible explanations. In these circumstances, review authors should attempt to understand the source of the small-study effects, and consider their implications in sensitivity analyses.