gms | German Medical Science

Background: Component network meta-analysis (CNMA) is an extension of standard network meta-analysis (NMA). It can be used when treatments are composed of common components, such as combinations of drugs or combinations of a drug with psychotherapy. In addition to knowledge of the structure of the network, we need a model to describe 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. We aim to investigate model selection in CNMA in a disconnected network.

Methods: We suggest two types of model selection strategies (forward and backward selection) based on the model fit, measured by the Q statistics, corresponding to the likelihood ratio statistic. We investigate and compare the strategies and demonstrate the methods on an example of multiple myeloma.

Results: CNMA models should be considered when encountering a disconnected network. They allow bridging gaps between separate parts of the network, if these have common components. The critical assumption of additivity should be explored. Though backward selection seems a particularly attractive way to mimic a standard NMA, we recommend to start with an additive model and add some interactions in a principled way, until a satisfactory fit is reached, preferably driven by subject-matter knowledge.

Conclusion: CNMA models, if feasible, are a promising alternative to matching approaches in disconnected networks. They are implemented in the R package netmeta.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.