Background: Mixed treatment comparison (MTC) models extend metaanalysis methods to enable comparisons to be made between all relevant comparators in the clinical area of interest. In such modelling, it is imperative that potential sources of variability are explored to explain both heterogeneity (variation in treatment effects between trials with pairwise contrasts) and inconsistency (variation in treatment effects between pairwise contrasts) to ensure the validity of the analysis. In order to allow for heterogeneity in treatment effects, a random effect is commonly included in evidence synthesis models, but this does not ensure inconsistency is addressed. Incorporation of study-level covariates allows systematic variability between trials to be explained which in turn can reduce both heterogeneity and inconsistency. Objectives: To extend the MTC framework to allow for the incorporation of study-level covariates in an attempt to explain between-study heterogeneity and reduce inconsistency. The models developed are applied to a 4-treatment network (consisting of 18 trials) for stroke prevention treatments in individuals with non-rheumatic Atrial Fibrillation. Methods: The MTC framework for binary outcomes outlined by Caldwell et al. (BMJ, 2005) is extended to include covariate x treatment interactions. Model (1) assumes that all treatment x covariate interactions are different and unrelated by including a separate regression coefficient for each treatment in the network; model (2) assumes that all treatment x covariate interactions are different for each treatment, but exchangeable; model (3) assumes that all treatment x covariate interactions are identical, implying the effect is the same regardless of treatment. Results: All three models are successfully applied to the 4-treatment network example with model (1) providing the best fit to the data. Note that, when the models are applied to the full 18-treatment network (consisting of 26 trials), the power to detect treatment by covariate interaction(s) is limited due to the scarcity of data, and only models (2) and (3) can be fitted. Conclusions: Here, we have demonstrated the feasibility of incorporating covariates to explain between-study heterogeneity and reduce inconsistency, although thought needs to be given regarding the appropriate model for the data available.