OBJECTIVES: Network meta-analysis (NMA) methods extend the pair-wise meta-analysis framework by allowing the simultaneous statistical synthesis of evidence on more than two interventions. Often the evidence base produces networks that are disconnected, such that there is neither direct nor indirect evidence to compare all relevant treatments within the network. Even when a fully connected network exists, the evidence base may present cross-trial variation in the reference arm outcomes, which, if not controlled for, may create biased estimates of relative treatment effect. These challenges are addressed using a case study on the biologic treatments for moderate to severe plaque psoriasis in children/young people as the motivating example.

METHODS: A series of Bayesian statistical NMA models were developed. The full set of relevant treatments could not be connected within the network of evidence such that only partial comparisons could be undertaken using conventional NMA approaches. We tackled this challenge by developing synthesis models which used external evidence from an adult population to connect the network in the children/young person population. Further model extensions were performed to adjust for population and cross-trial differences. Bayesian measures of fit and adequacy were used to compare and select models.

RESULTS: There were only 3 trials in children/young people in psoriasis, which formed a disconnected network of four biological treatments. External evidence from an adult population was used to bridge the evidence gap. Reference arm response rates were identified as a key source of heterogeneity. The NMA models were therefore adjusted for population and cross-trial differences providing an appropriate method to estimate relative treatment effects for all relevant treatments in children/young people.

CONCLUSIONS: Provided that valid external information can be inserted into the network, methodological approaches to NMA offer the opportunity to derive reliable estimates of relative treatment effects, while adjusting for key sources of heterogeneity.