OBJECTIVES: Network meta-analysis provides a robust methodology to obtain comparative estimates when head-to-head evidence is not available. However, it is always challenging to incorporate observational or single-arm studies in conventional NMA.

METHODS: A hypothetical situation reflecting the disconnection between four randomized controlled trials (RCTs) was simulated. Due to the lack of a connected network or a missing common comparator within the RCT evidence base, single-arm and observational studies were considered to generate a connected network. Analysis data (safety) and covariates for each treatment arm evaluated either in RCTs, or single-arm studies were simulated using the “wakefield” package in R. Single-arm studies and observational studies were matched to act as each other’s control group based on a distance metric derived from covariate information available in each study. Median age and proportion of males were the key covariates considered in the calculation for distance metric. A qualitative comparison of baseline characteristics was made to assess the comparability of studies included in the NMA. Bayesian NMA was conducted to determine the relative safety of treatment of interest versus other comparators. The three-level hierarchical model was considered to incorporate data from RCTs, single-arm trials, and observational studies while accounting for heterogeneity between the study designs. Due to considerable heterogeneity in the simulated studies, the random-effects model was considered.

RESULTS: Using the distance metric, four randomized controlled and 16 single-arm or observational studies were included for matching to bridge the gap between the disconnected networks. NMA results indicated that treatment of interest was associated with statistically significantly lower odds for the occurrence of generic adverse events when compared with other treatments.

CONCLUSIONS: The discussed NMA method can bridge the gap between the disconnected network of RCTs using single-arm and observational studies.