Introduction: Publication bias can distort or even invalidate the results of a systematic review. Amnesties for unpublished studies and prevention measures such as the prospective registration of RCTs offer no protection for non-randomised evidence included in systematic reviews. Further, this type of evidence may be victim to different discriminatory publication mechanisms from those existing for RCTs, such as discriminatory outcome reporting of results of data dredging exercises.

Objectives:

Methods: An empirical investigation was carried out to examine whether different levels of publication bias do exist within different study types. The randomised evidence for cholesterol-lowering treatments, and cohort studies examining the effect of cholesterol levels on coronary mortality were examined. Separate meta-analyses were carried out for each study type, funnel plots examined, and the method of 'Trim and Fill' (Duval and Tweedie 1999) applied. This method adjusts meta-analyses results for publication bias, in an attempt to quantify its influence on the different study types.

Results: The results of this will be fully and graphically described together with two analyses combining both study types, one ignoring, and one adjusting for publication bias.

Discussion: Finally, in light of these, consideration is given to measures that can be taken to deal with publication bias if both randomised and non-randomised evidence are to be included in a systematic review.