Table 4.

Comparison of frequentist and Bayesian approach

	Frequentist approach	Bayesian approach
NMA data input preparation	Input required for all pairwise comparisons per study (k (k − 1)/2), with k representing number of arms, equating to 1 for a 2-arm study, 3 for a 3-arm study, 6 for a 4-arm study etc	Input required for all comparisons to the baseline treatment, corresponding to number of arms − 1 (k − 1): 1 for a 2-arm study, 2 for a 3-arm study, 3 for a 4-arm study etc
Prior specification	No priors used – based solely on observed data	Prior distribution must be specified
Analysis	Straightforward implementation using available R package netmeta Statistical hypothesis testing is conducted. Output is presented as the estimates of effects and corresponding 95% CIs and associated p-values for the tests of significance	More computationally intensive, yet OpenBUGS code readily available at NICE Decision Support Unit No hypothesis testing takes place. Output is presented as the estimates of effect and corresponding 95% CrI, Bayes factor, and posterior probabilities of effect
Interpretation	Results are presented as estimated relative effects (mean difference or odds ratio) and 95% CIs, statistical significance can be determined Ranking of treatments through p-score, the frequentist equivalent to the SUCRA	Results are presented as summaries of the posterior distribution (of the mean difference or odds ratio) and 95% CrIs, no statistical significance can be determined. Results are interpreted as one treatment to be favorable/unfavorable over another treatment, or two treatments to be comparable Ranking of treatments through SUCRA. Probability of one treatment to be better than another treatment can additionally be estimated (not feasible in a frequentist approach)

CI confidence interval, CrI credible interval, NICE National Institute for Health and Care Excellence, NMA network meta-analysis, SUCRA surface under the cumulative ranking curve