Consistency between direct trial evidence and Bayesian Mixed Treatment Comparison: Is head-to-head evidence always more reliable?

Objectives This study aims to highlight the benefits of Bayesian mixed treatment comparison (MTC), within a case study of the efficacy of three treatments (pegfilgrastim, filgrastim and lenograstim) for the prevention of febrile neutropenia (FN) following chemotherapy.

Methods Two published meta-analyses have assessed the relative efficacy of the three treatments based on head-to-head trials. In the present study, all the trials from these meta-analyses were synthesised within a single network in a Bayesian MTC. Following a systematic review, the evidence base was then updated to include further recently-published trials. The metaanalyses and MTC were re-analysed using the updated evidence base.

Results Using data from the previously-published meta-analyses only, the relative risk of FN for pegfilgrastim vs. no treatment was estimated at 0.08 (95% confidence interval: 0.03, 0.18) from the head-to-head trial and 0.27 (95% credible interval: 0.12, 0.60) from the MTC, reflecting strong inconsistency between the results of the direct and indirect methodologies. When subsequently-published head-to-head trials were included, the meta-analysis estimate increased to 0.29 (95% confidence interval: 0.15, 0.55), while the MTC gave a relative risk of 0.34 (95% credible interval: 0.23, 0.54). The initial MTC results were therefore a better predictor of subsequent study results than was the direct trial. The MTC was also able to estimate the probability that there were clinically significant difference in efficacy between the treatments.

Conclusions Bayesian MTC provides clinically relevant information, including a measure of the consistency of direct and indirect evidence. Where inconsistency exists, it should not always be assumed that the direct evidence is more appropriate.