Suchergebnisse

Fujian Song, Lee Hooper, Yoon K LokeNorwich Medical School, University of East Anglia, Norwich, UKAbstract: Publication bias occurs when results of published studies are systematically different from results of unpublished studies. The term "dissemination bias" has also been recommended to describe all forms of biases in the research-dissemination process, including outcome-reporting bias, time-lag bias, gray-literature bias, full-publication bias, language bias, citation bias, and media-attention bias. We can measure publication bias by comparing the results of published and unpublished studies addressing the same question. Following up cohorts of studies from inception and comparing publication levels in studies with statistically significant or "positive" results suggested greater odds of formal publication in those with such results, compared to those without. Within reviews, funnel plots and related statistical methods can be used to indicate presence or absence of publication bias, although these can be unreliable in many circumstances. Methods of avoiding publication bias, by identifying and including unpublished outcomes and unpublished studies, are discussed and evaluated. These include searching without limiting by outcome, searching prospective trials registers, searching informal sources, including meeting abstracts and PhD theses, searching regulatory body websites, contacting authors of included studies, and contacting pharmaceutical or medical device companies for further studies. Adding unpublished studies often alters effect sizes, but may not always eliminate publication bias. The compulsory registration of all clinical trials at inception is an important move forward, but it can be difficult for reviewers to access data from unpublished studies located this way. Publication bias may be reduced by journals by publishing high-quality studies regardless of novelty or unexciting results, and by publishing protocols or full-study data sets. No single step can be relied upon to fully overcome the complex actions involved in publication bias, and a multipronged approach is required by researchers, patients, journal editors, peer reviewers, research sponsors, research ethics committees, and regulatory and legislation authorities.Keywords: publication bias, reporting bias, research-dissemination bias, evidence synthesis, systematic review, meta-analysis

The study protocol, publications, full study report detailing all analyses, and participant-level dataset constitute the main documentation of methods and results for health research. However, journal publications are available for only half of all studies and are plagued by selective reporting of methods and results. The protocol, full study report, and participant-level dataset are rarely available. The quality of information provided in study protocols and reports is variable and often incomplete. Inaccessibility of full information for the vast majority of studies wastes billions of dollars, introduces bias, and has a detrimental impact on patient care and research. To help improve this situation at a systemic level, three main actions are warranted. Firstly, it is important that academic institutions and funders reward investigators who fully disseminate their research protocols, reports, and participant-level datasets. Secondly, standards for the content of protocols, full study reports, and data sharing practices should be rigorously developed and adopted for all types of health research. Finally, journals, funders, sponsors, research ethics committees, regulators, and legislators should implement and enforce policies supporting study registration and availability of journal publications, full study reports, and participant-level datasets.