|
Abstract:
|
One of the challenges in network meta-analysis is how to effectively and efficiently integrate information when each of the studies only provides partial information for the multiple parameters (treatments). In this talk, we propose a general meta-analysis framework for analyzing multiple parameters. The framework applies to the network meta-analysis setting as well as the regular multivariate setting. The general idea is to combine multivariate confidence distribution (CD) functions, which can be viewed as frequentist "distribution estimates" of the unknown parameters. We show that the proposed CD framework yields 1) an efficient combination when the evidences are consistent; and 2) robust combinations when the population of the studies is contaminated. The properties of efficiency and robustness are illustrated using numerical examples.
|
