Home
Scholarly Works
Choosing the statistical model and between fixed...
Chapter

Choosing the statistical model and between fixed and random effects

Abstract

The choice of statistical model used in network meta-analysis (NMA) primarily depends on the type of outcome measured in the trials. In this chapter, we describe the unified generalized linear model (GLM) framework that can handle a wide range of outcomes, including those derived from binary, count, and continuous data types. We present both the fixed and random effects modelling approaches within the GLM framework, explain corresponding assumptions in the context of NMA, and discuss methodological issues and strategies on how to approach the decision between choosing fixed or random effects models. Along with the GLM framework, we present the Bayesian approach for inference, model fit assessment, and ranking treatments. Finally, we apply the Bayesian GLM models to three publicly available datasets, with binary, count, and continuous outcome, and demonstrate the use of deviance statistics to assess model fit for both fixed and random effect models.

Authors

Beyene J; Bonner A; Neupane B

Book title

Network Meta Analysis Evidence Synthesis with Mixed Treatment Comparison

Pagination

pp. 117-140

Publication Date

July 1, 2014

Contact the Experts team