What is Network Meta-Analysis (NMA)?
These comparisons create a web-like analysis called a Network Diagram or Network Comparison. Other names of NMA include Multiple Treatment Meta-Analysis, Mixed Treatments Comparison, Indirect Treatment Comparison, Pair-Wise Meta-Analysis and so forth. NMA has been widely used in technology appraisals for various clinical indications by technology assessment agencies around the world. For instance, in January 2019, the FDA performed a fixed-effect network meta-analysis to investigate risk of MACE associated with Romosozumab treatment.
Why NMA?
Meta-analyses of randomized controlled trials are considered the top of the hierarchy of clinical evidence. However, oftentimes, head-to-head comparisons are not available or are insufficient to answer a specific clinical question. NMA overcomes this limitation by providing a global estimate of efficacy or safety of multiple treatments that have limited or no direct comparisons. Furthermore, NMA allows for ranking of the treatments to obtain identification of the best option amongst all available options, provided that the statistical inference is valid. These appeals to clinicians and other decision makers as NMA can be used to answer the important question of “Which treatment is the best or worst?”
Methodological considerations in conducting NMA
Evidence synthesis
NMA requires a network of evidence which must be synthesized via a systematic and thorough literature search. The literature search also forms the basis for building the network in NMA, where each intervention/treatment is represented by a node and are linked if there is evidence of direct comparison in one or more studies. Indirect comparisons can be formed if two or more nodes share a common comparator when direct comparison between the nodes are unavailable. For instance, if treatment A is compared with placebo (C), and treatment B with placebo (C), then we can estimate the relative efficacy of A versus B via the common comparator C.
Statistical methods
Various modeling approaches can be applied to NMA, depending on the clinical question one aims to answer and nature of the data. The effect measurements can be modeled as either fixed or random, and the framework can be frequentist or Bayesian. In the case of observable between trial heterogeneity, transitivity may be achieved through inclusion of treatment-by-covariate interactions in the meta-regression model.
NMA can be performed in statistical software such as SAS, STATA, or R, and programs such as WinBUGS or JAGS, as well as several other software specifically designed for NMA (GeMTC, ITC).
Next Steps:
Read our analysis on other FDA Approval topics.
Schedule a one-on-one call to see how we can help with you FDA Submissions.
See how Princeton’s services can help you.
Subscribe to our blog to stay up to date.