Publication Bias: Assessment and Impact

Abstract

Publication bias is the systematic missingness of scientific results from the academic literature. In healthcare, the problem is sizable with nearly half of all studies that collect data never reaching full publication. This has the potential to result in systematic misdirection that can influence clinical decision making. The objectives of our study were three-fold. First we conducted a systematic literature review and meta-analysis to better understand what study characteristics influence their likelihood of publication and whether these interact with one another. Second, we evaluated whether this systematic misdirection resulted in a time differential from when meaningful evidence could have been available to when it actually reached researchers. The case study for this was the drug rosiglitazone. Finally, we developed a framework for evaluating the health and economic impacts of unpublished information. The results of our meta-analysis supported much of the previously published literature. A secondary analysis also highlighted an interaction between industry funded trials and result favorability. For our second objective, we found that for rosiglitazone, publication bias resulted in a 36-month differential between when meaningful evidence could have been available and when it was published. Currently available methods of adjustment were unable to mitigate this effect. Finally, for our last aim we developed a framework for assessing the impact of unpublished research and applied it to two case studies: rosiglitazone and rofecoxib. For rosiglitazone we found that this unpublished research resulted in significant wasted drug spend but no clinical events. For rofecoxib, publication bias was found not to affect decision making.