Two decades after its introduction into clinical medicine, meta- analysis has become established as a fundamental tool of evidence-based practice. Still, key methodological issues remain. One issue frequently raised by both advocates and critics is the problem of publication bias, which can threaten the results of any meta-analysis. Medical studies that achieve statistical significance are more likely to be published than those that do not, and among published studies, those that report a positive effect tend to be published sooner than negative studies. This phenomenon, called publication bias, may lead to erroneous conclusions that could seriously impact payers, policy makers, and patients. The AHCPR's Evidence-Based Practice initiative and the Cochrane Collaborative rely on meta-analyses of randomized clinical trials to make recommendations for clinical practice and health care policy, and the FDA accepts meta-analyses for secondary indications. A number of methods have been developed to detect and adjust for selection bias (any bias influencing the retrieval of studies, including publication bias). There has not yet been a comprehensive evaluation and comparison of the methods, and the techniques have yet to be applied to a large number of meta- analyses, to estimate the impact of selection bias on quantitative reviews in journals and databases. The goal of this project is to address these critical gaps, as well as to develop and test new methodology for detecting and adjusting for bias.
