Article type
Year
Abstract
Background: Network Meta-Analysis (NMA) pools direct and indirect evidence on relative treatment effects in a simultaneous analysis. NMA is increasingly applied in the evaluation of multiple competing treatments, however it requires substantially more systematic review resources. One frequently asked question is ‘how far should I extend the network?’
Objectives: To assess the benefit, in terms of precision of effect estimates for AvB, from extending the network.
Methods: We consider hypothetical networks with six-treatments, beginning with a ‘star-shaped’ network. We then increase the complexity of the network, first to form an ABC triangle, and gradually add more evidence connecting treatments in the network. We explore the effects of varying the relative strength of direct compared to indirect evidence, and the impact of heterogeneity.
Results: In all scenarios extending the network increased precision of the AB treatment effect. Under a fixed effect model, gain in precision was modest (53% increase) when the existing direct AB evidence was already strong and was substantial when the direct evidence was weak (1143%). Under a random effects model, gain in precision was lower when heterogeneitywas high (449% gain with τ2 = 1 compared with 933% gain when τ2 = 0.1). In all scenarios there was a ‘ceiling effect’ beyond which additional indirect evidence did not contribute increased precision for the AB estimate. This occurred when the AB comparison was fully connected with all other treatments (although not a complete network).
Conclusions: Including additional treatments in a connected network increases precision. However, the gain in precision is only modest if the direct evidence is already strong, or if there is a high degree of heterogeneity. Once the focal treatment comparisons of interest have been connected to all of treatments that have been trialled for the indication, there is no additional benefit in extracting further information.
Objectives: To assess the benefit, in terms of precision of effect estimates for AvB, from extending the network.
Methods: We consider hypothetical networks with six-treatments, beginning with a ‘star-shaped’ network. We then increase the complexity of the network, first to form an ABC triangle, and gradually add more evidence connecting treatments in the network. We explore the effects of varying the relative strength of direct compared to indirect evidence, and the impact of heterogeneity.
Results: In all scenarios extending the network increased precision of the AB treatment effect. Under a fixed effect model, gain in precision was modest (53% increase) when the existing direct AB evidence was already strong and was substantial when the direct evidence was weak (1143%). Under a random effects model, gain in precision was lower when heterogeneitywas high (449% gain with τ2 = 1 compared with 933% gain when τ2 = 0.1). In all scenarios there was a ‘ceiling effect’ beyond which additional indirect evidence did not contribute increased precision for the AB estimate. This occurred when the AB comparison was fully connected with all other treatments (although not a complete network).
Conclusions: Including additional treatments in a connected network increases precision. However, the gain in precision is only modest if the direct evidence is already strong, or if there is a high degree of heterogeneity. Once the focal treatment comparisons of interest have been connected to all of treatments that have been trialled for the indication, there is no additional benefit in extracting further information.