Article type
Year
Abstract
Background: Until recently, network meta-analyses (NMA) failed to address the certainty associated with different paired comparisons in a network. The GRADE working group has developed an approach to NMA that involves rating the certainty in estimates for each paired comparison.
Objective: To illustrate GRADE’s approach to NMA.
Method: The GRADE approach was developed through discussion and iterative refinement; we have applied the approach in a number of NMAs. The approach involves four steps: 1) present direct and indirect treatment estimates for each paired comparison; 2) rate the certainty of each direct and indirect estimate; 3) present the NMA estimate for each comparison; 4) rate the certainty of each NMA estimate.
Results: Rating the certainty of each direct estimate is straightforward using established GRADE methodology. Rating the certainty for indirect estimates involves identifying, for each paired comparison, the 1st order loop (those loops that involve only a single additional intervention linked to the two treatments/nodes of interest) that contributes most to the indirect estimate; rating the certainty for the two direct comparisons contributing to the indirect comparison; and choosing the lower of the two ratings. If direct and indirect estimates are consistent, the higher rating of the two represents the overall certainty of the network estimate. If they are inconsistent, GRADE suggests using the higher of the two, rather than the network estimate, as the best estimate of effect. Examples illustrate the utility of the approach. For instance, in a NMA addressing optimal fluid administration in septic patients, high certainty evidence was available for the comparison of starch vs crystalloid, but only very low certainty for gelatin vs crystalloid. A challenge to full application of the GRADE approach is difficulty in generating separate indirect effect estimates in NMAs with large numbers of comparisons.
Conclusions: Since clinicians need to take into account not only best estimates, but also certainty, in making their management decisions, the GRADE approach enhances the utility and interpretation of NMA.
Objective: To illustrate GRADE’s approach to NMA.
Method: The GRADE approach was developed through discussion and iterative refinement; we have applied the approach in a number of NMAs. The approach involves four steps: 1) present direct and indirect treatment estimates for each paired comparison; 2) rate the certainty of each direct and indirect estimate; 3) present the NMA estimate for each comparison; 4) rate the certainty of each NMA estimate.
Results: Rating the certainty of each direct estimate is straightforward using established GRADE methodology. Rating the certainty for indirect estimates involves identifying, for each paired comparison, the 1st order loop (those loops that involve only a single additional intervention linked to the two treatments/nodes of interest) that contributes most to the indirect estimate; rating the certainty for the two direct comparisons contributing to the indirect comparison; and choosing the lower of the two ratings. If direct and indirect estimates are consistent, the higher rating of the two represents the overall certainty of the network estimate. If they are inconsistent, GRADE suggests using the higher of the two, rather than the network estimate, as the best estimate of effect. Examples illustrate the utility of the approach. For instance, in a NMA addressing optimal fluid administration in septic patients, high certainty evidence was available for the comparison of starch vs crystalloid, but only very low certainty for gelatin vs crystalloid. A challenge to full application of the GRADE approach is difficulty in generating separate indirect effect estimates in NMAs with large numbers of comparisons.
Conclusions: Since clinicians need to take into account not only best estimates, but also certainty, in making their management decisions, the GRADE approach enhances the utility and interpretation of NMA.