DIA/FDA Statistics Forum 2014

This session will have a particular focus on "network meta-analysis" for assessment of product safety. Standard meta-analysis, whether of continuous or discrete outcomes, has focused exclusively on comparing two treatments or classes of treatments. Recently, methods have been developed to integrate comparisons of multiple treatments into coherent models that allow simultaneous comparison of all treatments. Applications of these methods are beginning to become popular in the clinical literature. Called variously network meta-analysis, multiple treatments meta-analysis and mixed treatment comparisons, the basic idea is that the direct evidence from head-to-head trials may be combined with the indirect evidence from trials that involve common comparators. For instance, trials of A vs. B and A vs. C give indirect information about the comparison of B vs. C under the assumption of consistency that the direct and indirect comparisons estimate the same quantity. Our speaker will discuss basic principles of networks and their analysis including the assumptions of consistency and treatment of heterogeneity and will discuss the validity of these assumptions in the context of safety studies relative to efficacy studies. Some important features of safety data to be discussed include length of follow-up, rare events, varying patient populations, differing exposures, variation in medical practice, age of treatments, trial sponsor. Extensions to multiple outcomes and longitudinal designs with incomplete data will also be discussed. A panel will discuss questions related to meta-analysis, network meta-analysis and the FDA Public Meeting on meta-analysis held November 2013.