Network meta-analysis of biological response modifiers in rheumatoid arthritis including real world evidence at multiple time points

Objectives: Network meta-analysis (NMA) is widely used to compare multiple interventions of interest when head-to-head comparisons of active treatments are not available. Most NMAs pool data from randomised controlled trials (RCTs) on a single clinical outcome. However, in the case of chronic diseases such as rheumatoid arthritis (RA), outcomes are often reported at different time points and long-term real-world data (RWD) is routinely collected as part of national registries. Evaluation of models for the inclusion of different time measures in NMA, especially from both a regulatory and reimbursement perspective, is thus warranted and is considered here. Methods: RCTs and observational studies evaluating biological agents in RA were searched using standard filters and electronic databases. Networks of RCTs were supplemented with RWD to include outcomes extracted for as many time points as possible. Multivariate NMA models were extended to incorporate repeated measures, adjusting for correlation between time points and bias of RWD. Sensitivity and scenario analyses were performed to test different network sizes, correlation structures and bias adjustments. Results: Addition of RWD and studies reporting treatment effects at multiple time points significantly increased the evidence base for NMA in RA. The inclusion of RWD led to a reduction in the level of uncertainty around most of the effect estimates. Furthermore, the additional evidence from multiple times has potential of reducing uncertainty by 'borrowing' evidence and giving a fuller view of treatment effect over time, not just at a specific single time point. Conclusions: Initial evaluation of these models in NMA indicates that extending an evidence base to include repeated measures and RWD maximises study network sizes and can significantly impact the level of uncertainty in treatment effects. Further investigation of correlation and bias modelling is warranted, as too is the application of new NMA fractional polynomials model to RA.