OBJECTIVES: Particularly in the context of HTA evaluations where both post-marketing and pre-marketing data may be considered, the evidence to be synthesized can be sparse, partial and heterogeneous for safety outcomes. The Bayesian option has increasingly appeared as an unrivalled option for such challenging evidence synthesis cases but implementation in practice may be questioned. This work aims at determining how Bayesian meta-analysis or mixed treatment comparison of safety data can be optimized especially regarding the choice of prior distributions and model parameterization.  METHODS: Based on the latest developments from the DIA working group on Bayesian methods for safety data applied to specific real-world cases of both direct meta-analysis and mixed treatment comparisons (MTC), different model parameterizations and different forms of informative and non-informative prior distributions are tested, with various weights allocated to the clinical data vs. the observational information. RESULTS: As opposed to the NICE parameterization of network meta-analysis, the 2-way predictor parameterization of MTC as proposed by the DIA working group provides more robust analysis based on non-informative priors. In the case of informative prior results, the most robust option was seen for equal total weight of clinical vs. observational data. Results of all meta-analyses appeared to be consistent across different model and prior specifications, even with low number of studies (<10). CONCLUSIONS: Bayesian evidence synthesis can leverage all available information in a robust manner for both direct and indirect comparisons, with fair quantification of uncertainty. Specific guidance on MTC model parameterization for safety data could complement the current NICE guidelines.