Table 5 Summary of key inputs and outputs
Frequentist approach | Bayesian approach | |
|---|---|---|
Is the network of evidence sparse? (<5 studies) | Works well | Does not work well for standard non-informative priors but works well for informative priors Non-informative priors could result in unrealistically wide credible intervals |
Is prior specification justified? | No priors used—based solely on observed data | Non-informative if enough data are available, informative in the case of sparse data, choice of suitable distributions and additional information such as expert clinician opinion |
Are there few large studies of high quality? | Consider FE model | Consider FE model |
Are there country-specific regulations? | Required by German G-BA and Australian PBAC | Preferred by NICE in the UK |
Interpretation | Statistical significance or the absence thereof | One treatment favorable/unfavorable over another treatment, or two treatments comparable Often falsely interpreted as significant or not significant; common misconception |