cochrane collaboration > colloquia > abstracts > ottawa 2004 Bayesian statistics in meta-analysis and meta-regression

Jill Hayden, George Tomlinson

Objectives: The purpose of this training workshop is to provide systematic reviewers and methodologists with insight into the practical and conceptual advantages of incorporating Bayesian statistical methods into meta-analysis. Potential benefits are demonstrated and discussed in the context of a recently completed systematic review. Description: Systematic reviews may include individual studies with data that present challenges for appropriate analysis using classical statistical methods (e.g. a review may include studies with both single and multiple comparison groups, a single follow-up or multiple follow-ups, and variations on similar interventions). The Bayesian statistical approach to meta-analysis has been recommended and recently the use of Bayesian methods has been increasing. Using practical illustrations and motivating the methods through wish-lists elicited from participants, this workshop will discuss advantages of Bayesian methods for meta-analysis and meta-regression. These include: 1) More complete representation of between-study heterogeneity; 2) More transparent and intuitive reporting of results, including: a. direct probability statements about clinically meaningful effects; b. prediction of future treatment effects; 3)The ability to include objective prior information where it exists and to show how sensitive the results are to different subjective prior opinions; 4)Ease of specification of realistically complex models, which in turn allows: a. ranking of different levels of intervention characteristics; b. use of all the relevant data in one model; c. indirect and direct covariate comparisons in same model; d. prediction of results for the best combination of characteristics.

At the end of the workshop, participants will be aware of the potential advantages of using Bayesian meta-analytic methods. Basic programs for common meta-analytic models, written in the WinBUGS language, will be provided to participants.

Ottawa 2004 W-065