Skip to main content

Languages

Implementing the extended risk of bias tool for non-randomised studies: feedback from the Cochrane Epilepsy Group

Date and Location

Session: 

P2.016

Date

Saturday 21 September 2013 - 10:30 - 12:00

Location

Presenting author and contact person

Presenting author

Jennifer Pulman

Contact person

Jennifer Pulman
Abstract text
Background: Over the past couple of years the Cochrane Non-Randomised Studies Methods Group have been developing the Cochrane risk of bias tool to be used with non-randomised studies. This has included the addition of a seventh domain examining the aspect of confounding variables and the implementation of a scale ranging from 1 (low risk of bias) to 5 (high risk of bias). The Cochrane Epilepsy Group (CEG) tested the extended tool in two systematic reviews that included non-randomised studies, further results to follow from a third review. Methods: A reference table of 1-5 scale parameters was developed by the CEG to aid decision making on each domain of bias. 46 non-randomised studies were assessed using the extended risk of bias tool from two systematic reviews. For allocation concealment and sequence generation each study was rated either low/high or unclear. For blinding, confounding variables, incomplete outcomes data, selective reporting and other bias each study was rated 1-5 according to pre-specified criteria. Results: For allocation concealment and sequence generation all 46 studies were rated as high risk of bias and agreement was 100%. For blinding, agreement was reached +/-1 in 45 studies. For confounding variables agreement was reached +/-1 in 43 studies. For incomplete outcome data, agreement was reached +/-1 in 38 studies. For selective reporting agreement was reached +/-1 in 44 studies, and for other bias agreement was reached +/-1 in 36 studies. Using the tool was time-consuming with regards to developing the 1-5 scale, training other authors and first usage however became more feasible with time. Conclusion: Agreement on bias judgements was more prevalent in the domains of allocation concealment, sequence gerneration, blinding and confounding variables. Challenges with using the extended risk of bias tool and recommendations will be presented.