Exploring the applicability and adaptation of the GRADE system to results from network meta-analysis: a pilot study

Article type
Authors
Del Giovane C1, Chaimani A2, Caldwell D3, Salanti G2
1Italian Cochrane Centre, Modena, Italy
2Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Ioannina Greece
3School of Social and Community Medicine, University of Bristol, UK
Abstract
Background: The GRADE approach provides an explicit and transparent system for rating the quality of evidence frommeta-analysis. Five components are considered: study limitations, inconsistency, indirectness, imprecision and publication bias. However, the evaluation of these five components across the body of evidence feeding into Network Meta-Analysis (NMA) and its results may not be straightforward and may require adaption.

Objectives: To explore the ease of adapting the GRADE approach to the output from NMA.

Methods: A random sample of five NMAs was identified from different medical fields to evaluate the suitability of using GRADE. For each component, two independent reviewers evaluated whether the standard GRADE criteria for judging the evidence could be applied. Where this was not possible for both reviewers and any difficulties were not resolved by a third one, we proposed modifications and presented graphical tools that can be used to facilitate judgment.

Results: Adaptation of current guidance for judging the GRADE items is needed in the context of network meta-analysis. There is no straightforward way to summarize the study limitations across a network and judgments about the presence and the role of publication bias are not easy to derive. Additional item(s) to reflect the disagreement between sources of evidence need to be added.

Conclusions: The standard GRADE system is not sufficient for assessing the evidence base in a NMA as important uncertainties exist regarding the role of bias and the importance of indirectness. The criteria need adaptation and a specific guidance to network meta-analysis needs to be produced.