Reviews

Principal investigator

Dr. Guido Schwarzer
Institute for Medical Biometry and Medical Informatics
University Hospital Freiburg
Stefan-Meier-Str. 26, 79104 Freiburg, Germany
Phone: ++49 (0)761 203 6668
Fax: ++49 (0)761 203 6680

Researchers

Summary

The overarching project aim was to review, evaluate and develop `modern statistical methods for the detection, quantification and correction of bias in the results of systematic reviews', taking the Copas model for publication bias as a starting point.
A suite of R functions to fit the Copas model, then tabulate and graph the results, has been developed and made available as part of the meta package. We utilised this R package to evaluate the practical utility of the Copas model both by simulation and by analysing 160 meta-analyses with binary outcomes.
We have proposed a new model for publication bias which we believe is more accessible to Cochrane reviewers than the Copas model. Further, we have proposed a rapid, approximate method for fitting this model, which utilises an approach developed by one of us for sensitivity analysis after multiple imputation. Furthermore, we have proposed a new test for detecting publication bias, and evaluated this and all existing tests using a comprehensive simulation study. This evaluation showed our new test is one of the most accurate and powerful.
Finally, we have undertaken an empirical study on the relationship between (a) the p-value of the primary hypothesis test reported by a study and (b) the number of citations the study subsequently receives. This showed that, even after adjustment, articles reporting small p-values were cited more frequently.
Our proposal for the next funding period builds directly on this work, and will naturally lead to collaborations with the Non-randomized and Pharmacoepi groups. In particular, the new multiple-imputation based method for publication bias provides a nice framework to approach the problem of selection bias in non-randomized studies. Furthermore, we want to explore whether a meta-analytic approach can be exploited to simplify the analysis of large complex international observational studies.

Publications