Article type
Year
Abstract
Background: Several immunotherapies exist for the management of multiple sclerosis and their relative efficacy can be evaluated via Multiple-Treatments Meta-analysis (MTM). Some agents are administered at different doses, but the impact of increasing the dose on delaying the disease progression is still unclear. Therefore it is uncertain whether different doses of the same agent can be considered to form a common 'node’ in the treatment network.
Objectives: a)to explore a series of MTM models where the uncertainty about the impact of dose is embedded in the definition of each node in the treatment network b) to evaluate the assumptions that dose effects are the same, randomly different but exchangeable or follow a monotonic pattern.
Methods: We developed Bayesian models considering the different node definitions. We started with the network where each agent, whatever the dose, defines a node. We gradually 'blurred’ the nodes so that different dose effects of the same agent are: random but exchangeable; isotonic or have a linear pattern around the agent’s mean effect. We ended up with the network where each different dose defines an independent node. Each model has different implications for the assumption of consistency of effect (either at agent level or at dose level). The models were compared using goodness of fit criteria; changes in heterogeneity and inconsistency were monitored.
Results: Models for linear and independent dose effects had poor fit whereas exchangeable dose effects models had the best fit and lowest heterogeneity. This suggests that the dose effects of the same agent approximate the mean agent effect and hence a dose-response association is unlikely.
Conclusions: The 'node-blurring’ approach can be a useful exploratory tool when there is doubt whether similar interventions should be grouped under the same node in a network; a dilemma which is frequently encountered in MTM.
Objectives: a)to explore a series of MTM models where the uncertainty about the impact of dose is embedded in the definition of each node in the treatment network b) to evaluate the assumptions that dose effects are the same, randomly different but exchangeable or follow a monotonic pattern.
Methods: We developed Bayesian models considering the different node definitions. We started with the network where each agent, whatever the dose, defines a node. We gradually 'blurred’ the nodes so that different dose effects of the same agent are: random but exchangeable; isotonic or have a linear pattern around the agent’s mean effect. We ended up with the network where each different dose defines an independent node. Each model has different implications for the assumption of consistency of effect (either at agent level or at dose level). The models were compared using goodness of fit criteria; changes in heterogeneity and inconsistency were monitored.
Results: Models for linear and independent dose effects had poor fit whereas exchangeable dose effects models had the best fit and lowest heterogeneity. This suggests that the dose effects of the same agent approximate the mean agent effect and hence a dose-response association is unlikely.
Conclusions: The 'node-blurring’ approach can be a useful exploratory tool when there is doubt whether similar interventions should be grouped under the same node in a network; a dilemma which is frequently encountered in MTM.