A confidence distribution approach for an efficient network

This paper presents a new approach for network meta-analysis that combines multivariate con9 fidence distributions (CDs). Network meta-analysis generalizes the traditional meta-analysis of 10 pairwise comparisons to synthesizing studies for multiple treatment comparisons, and supports 11 inference on all treatments in the network simultaneously. It can often strengthen inference on 12 a pairwise comparison by borrowing evidence from other comparisons in the network. Current 13 network meta-analysis approaches are derived from either traditional pairwise meta-analysis 14 or hierarchical Bayesian methods. This paper introduces a general frequentist approach for 15 network meta-analysis by combining CDs, which are viewed as frequentist “distribution estima16 tors”. Instead of combining point estimators, the proposed approach combines CD functions, 17 which contain richer information, and thus yields greater efficiency in its inferences. This paper 18 shows that the proposed CD approach can efficiently integrate all the studies in the network 19 even when individual studies provide comparisons for only some of the treatments. Numerical 20 studies, through real and simulated data sets, show that the proposed CD approach generally 21 outperforms traditional pairwise meta-analysis and a commonly used Bayesian hierarchical 22 model. Although the Bayesian approach may yield comparable results with a suitably chosen 23 prior, it is sensitive to the choice of prior, which is often subjective. The CD approach is 24 prior-free and can always provide a proper inference for the treatment effects regardless of the 25 between-trial covariance structure. 26