Assessing Consistency in a Network Meta-Analysis to Compare Once Weekly Dulaglutide Versus Other Glp-1 Receptor Agonists in Patients with Type 2 Diabetes.

Abstract

 The ‘node-splitting’ model (Dias et al. 2010)  The direct and indirect evidence for each pairwise contrast are split out and compared  Model assesses whether relaxing the consistency constraint improves model fit  The ‘inconsistency’ model (Dias et al. 2010; Dias et al. 2013)  Independent treatment effects, whereby the consistency constraint is relaxed, are estimated for each treatment comparison for which trial evidence is available.  Where trials have more than two treatment arms, independent treatment effects are estimated for the comparisons against a common reference arm rather than for all the possible comparisons within a trial.  This can be compared to the ‘consistency’ model which is the traditional NMA approach assuming consistency between studies in the treatment effects.  The ‘design-by-treatment interaction’ model (Higgins et al. 2012).  ‘design’ refers to the combination of treatments included in a study e.g. estimating the relative treatment effect of A vs B using different designs AB (two arm study comparing A and B) or ABC (3 arm study comparing A, B and C).  The inconsistency between treatment effects using alternative study designs is estimated (Higgins et al. 2012)  These models were applied to the datasets to determine whether inconsistency was present and to examine potential causes. Table 2: Inconsistency results from ‘design-by-treatment’ model (16-36 week add-on to MET ± SU ± TZD) Assessing consistency in a network meta-analysis to compare once weekly dulaglutide versus other GLP-1 receptor agonists in patients with type 2 diabetes