OBJECTIVES: The Bayesian and the Frequentist methods are two different approaches for performing a network meta-analysis (NMA). The choice of the method is mostly based on preference of the statistician (Bayesian or Frequentist school) and not on a specific rationale. The objective of this study is to replicate published NMAs that have used one of the two approaches, conducting the NMA with the other approach and determining possible differences in the outcomes.

METHODS: Two published NMAs with complete information of the underlying methodology and the statistical package in R and Stata were selected. NMAs selected include one on Relapsing Remitting Multiple Sclerosis (RRMS) with R package (Huisman et al., 2017) and another on Plaque Psoriasis (PP) with Stata package (Sbidian et al., 2017). Outcomes for risk ratio (RR) and rate ratio, p-score and SUCRA were used for the comparison. For the Frequentist method, netmeta package for R was used. Input data was taken from original publications. The Frequentist NMAs were run with R-Studio version 1.1.

RESULTS:

RRMS NMA: 16 treatments were analyzed in RRMS. For the rate ratios, no differences were observed in 8 treatments, while the difference ranged from 0.025 to 0.006 in the remaining 8 treatments. The comparison between SUCRA and p-score was identical in 15 treatments and the difference was 0.01 in 1 treatment.

PP NMA: 20 treatments were assessed for PASI 90 outcome. For RR, no difference was found in 4 treatments with a mean relative difference of -0.1% for all treatment comparisons. Differences for SUCRA vs p-score were similar.

CONCLUSIONS: The Bayesian and Frequentist approaches used for NMAs are similar for rate ratios and the ranking. Differences seem more related to the package used (Stata or R) than the underlying methods. Further research is necessary to investigate this.