Network meta-analysis (NMA) is a generalization of pairwise meta-analysis, so far mainly used in heath care applications. A network consists of multiple interventions - that form the the nodes - and comparisons between them, based on studies, corresponding to the edges of the network. NMA allows estimating relative effects of all pairs of interventions in the network by combining direct and indirect evidence. Component network meta-analysis (CNMA) is an extension of standard NMA. It can be used when treatments are composed of common components, such as combinations of drugs or combinations of a drug with a psychotherapy. In addition to knowledge of the structure of the network, a model is needed that describes how the effects of treatment components add in combination. The simplest and most parsimonious model is an additive model, which can be enriched by adding interaction terms. A special property of CNMA models is that they can be applied also in cases where the network is disconnected, where standard NMA is impossible, provided that all subnetworks have at least one common component. One has to choose whether to apply the sparsest (that is, the additive) model or to add a number of interactions, and how many. The more interactions are added, the more the model resembles a standard NMA. We suggest a model selection strategies (forward selection) based on the model fit, measured by the Q statistics that is popular in meta-analysis. I will demonstrate the methods on an example treatments for depression.