OBJECTIVES: To investigate the effect of omitting treatments from network meta-analyses on overall treatment effects and treatment rankings. METHODS: We selected published network meta-analyses that met the following criteria: compared t≥5 treatments, had ≥2 loops, ≥2tstudies and set to determine treatment superiority. If multiple published analyses considered the same treatments (e.g. multiple networks pertaining to COPD drugs), the larger network was selected. We defined a node’s connectivity as its number of edges. Each network was analyzed systematically with the removal of one node at a time. Nodes that were in ≥50% of studies were not removed. Impact of node exclusion was measured using the relative change in treatment effect estimates, changes in the top-three ranked treatments, and changes in probabilities of being the best treatment. Relative changes in effect size were expressed as fold-deviations. For each network with excluded node(s), we measured the maximum and geometric mean of fold-changes. RESULTS: In total, 19 networks were selected for analysis. Approximately half the networks had average fold-change larger than 1.10 (greater than 10% relative change in treatment effects). Approximately half of the networks also had changes in the top three ranks and substantial changes in treatment rank probabilities. Within these networks, the maximum fold-change was generally larger than 1.25. In networks with no changes in top-three ranked treatments, the ‘best’ treatment mostly had probability ≥70% of being the best. Two features were consistent across the nodes leading to the largest change in probabilities and effects: they were among the most connected nodes and tended to have a 0% probability of being the best treatment. CONCLUSIONS: Network meta-analytic methods are still in their infancy. Our results suggest that failing to include one or more treatments within a network can lead to important changes in conclusions reached.