Background: Traditional meta-analyses only include direct head-to-head studies. However, it may be unethical to compare a novel treatment to placebo if an effective treatment exists. A meta-analysis restricted to direct comparisons would require many years to evaluate a novel treatment in the context of overall patient care. Mixed-treatment meta-analysis (MTM) overcomes these limitations by including indirect evidence, i.e. compare treatment A to treatment B when there are studies comparing each to placebo. Current methods for MTM use only the variance data to combine studies and ignore other inter-study differences that might affect the effect estimate, except in an adhoc analysis.

Objectives: To describe an approach to mixed-treatment meta-analysis that allows the incorporation of study characteristics that potentially modify the estimate of the effect.

Methods: Any meta-analysis is simply an observational study where each study represents an individual subject, and MTM can be viewed as a missing data problem. For example, study A has the 'subject' characteristics: year 1995, 35% female, OR (A vs Ref) = 2.1 (95%CI = 1.2, 3.9), and a 20% reference group mortality. Study B's characteristics are: year 1998, 30% female, OR (B vs Ref) = 0.9 (95%CI = 0.5, 1.6), and a 30% reference group mortality. Comparing these two studies, study A has missing data for the variable "B vs Ref" and study B has missing data for the variable "A vs Ref". We use all the available data and the multiple imputation method to obtain estimates for the missing values. We use data on migraine treatments to compare the results from traditional MTM to our approach.

Results: The data comparing the incoherence observed between the estimate of effect from the two approaches are currently being analysed and will be presented.

Conclusions: A novel approach for MTM is described. Validation of the approach is in progress.