Bayesian approaches to analysing adverse events in randomised trials with a high proportion of trial arms observing no event
Research Objectives:
When analysing a clinical trial with binary endpoints, several effect measures are of interest: risk difference, risk ratio, and odds ratio. Meta-analyses on adverse events often include trials in which zero events were observed in one or both treatment arms. For these trials neither risk ratio nor odds ratio is defined. Various solutions have been proposed, e.g. the use of continuity corrections or the arcsine difference. It has been claimed, that trials with zero events in both arms provide no information on risk ratio or odds ratio.
Description of work:
We propose to take a fully Bayesian approach for the meta-analysis of such trials, simultaneously modelling all the relevant parameters, the baseline risks in control arms and the treatment effect. Sensitivity to the choice of the necessary prior distributions is important and will be explored in simulation studies. Furthermore, predictive simulations will be done to assess whether predictive model checks will improve when available prior knowledge on baseline risks in the treatment arms is formally incorporated in the analysis.
Host Institution: University of Bern