Illustrating the Assumptions of Meta-Regression in Treatment Networks

Background Network meta-analysis (NMA) is a common statistical method used to synthesize evidence across multiple studies, enabling simultaneous comparison of multiple competing treatments for a given condition. Network meta-regression (NMR) extends NMA by adjusting treatment effect estimates based on study-level characteristics, helping to explain residual heterogeneity. Despite its usefulness, NMR adoption as a technique has been limited due to conceptual complexity, implementation challenges, and limitations in sparse-data settings. NMR model coefficients follow independent, exchangeable, or common assumptions, each applied with or without enforcing consistency among treatment comparisons. However, choosing between these models is complex, as different modeling assumptions can lead to varying results and interpretations depending on factors like model fit and data availability. Additionally, before fitting NMR models without consistency, meta-analysts must also consider potential data directionality, where certain study characteristics (e.g., sponsorship) may systematically bias treatment effect estimates. Methods We present frequentist tools for NMR and evaluate the properties of four different modelling assumptions achievable via design matrix modifications to the standard NMA model. Based on these properties, we provide recommendations for NMR implementation that account for data availability and the research question of interest. We also offer guidance on interpreting treatment-by-covariate interactions in the context of underlying NMR assumptions. Results We illustrate the different NMR modeling assumptions using a network of ten diabetes treatments. Findings highlight the impact of data directionality in models without consistency in the treatment-by-covariate interactions. Differences across modeling assumptions underscore the importance of carefully considering both the network structure and the characteristics of included studies when implementing NMR. Conclusion This work elucidates crucial NMR assumptions and demonstrates the importance of the network structure for NMR. The introduced tools can streamline the implementation of NMR, facilitating the exploration of sources of heterogeneity and inconsistency in NMA and expanding tools available to researchers in the field of evidence synthesis.