Background: Bayesian meta-analysis has gained popularity in the field of evidence synthesis. Public health interventions are targeted to highly heterogeneous populations, multi-component interventions, multiple outcomes, influenced by context and, moreover, the answer to a question comes from a hierarchy of studies, which is in contrast to the clinical interventions. In such situation Bayesian methods have a wide scope of application.
Objectives: To develop methods to incorporate Bayesian methods into public health evidence synthesis.
Methods: Public health systematic reviews were identified which included studies with different type of study designs for a single question. Data were extracted from those studies for the outcome(s) concerned. Bayesian meta-analysis was performed with likelihood from the stronger studies and the results of the weaker studies as prior. The analysis was performed in WINBUGS software with 5000 iterations for burning the sampler and 10,000 iterations for estimation. Convergence was assessed using autocorrelation and density plot. A simulation study was also performed with different meta-analysis situations with repetitions of 1000 in each case.
Results: The concept has been applied in ten public health systematic reviews, which included different study designs to answer the same question. Different combinations of priors and likelihood were drawn and odds ratio, relative risk with 95% CrI, CI were calculated. The results were then compared with the estimates of traditional meta-analysis with the posterior estimates of Bayesian.
Conclusions: The present paper describes a mechanism to incorporate all the levels of a body of evidence for a single question using the Bayesian approach.