OBJECTIVES: The assumption of transitivity is central to network meta-analysis (NMA): clinical studies comparing treatments A and B must be sufficiently similar to those comparing B and C for an indirect comparison between A and C to be valid. The practice of lumping similar treatments into a single node (eg, grouping drug classes; grouping doses) could bias analyses; however, splitting treatments into independent nodes can add complexity or disconnectedness to network structure. As a case study, we evaluated theoretical and practical considerations relating to handling platinum doublets (PD) as lumped or split nodes in an NMA of advNSCLC therapies. METHODS: We conducted a systematic literature review in MEDLINE^®, Embase, and published technology appraisal guidance from the National Institute for Health and Care Excellence to identify NMAs in advNSCLC that included PD. We extracted data on network structure, endpoints, relative effects, and rationale for lumping versus splitting. We evaluated the data against a theoretical framework of methodological and clinical considerations for lumping versus splitting. RESULTS: From 136 hits, we identified 14 NMAs that included PD as comparators in an advNSCLC network; seven lumped PD, five split PD, and two partially split PD. Among split networks, relative efficacy was similar across most PD; however, significant differences were evident in safety endpoints. The outcome-specific application of our framework compared the magnitude and direction of bias introduced by lumping certain PD against the changes in network structure imposed by splitting. Our framework highlighted the potential to introduce exchangeable effects for certain PD as an intermediate between lumping and splitting, obviating certain challenges of splitting while avoiding the strong lumping assumption. CONCLUSIONS: Carefully defining NMA treatment nodes is a critical step toward producing methodologically rigorous and clinically meaningful results. This exploration applies to a broad range of researchers faced with the decision of whether to lump or split nodes.