A Bayesian ordinal model for heterogeneity in a multi-centre myocardial infarction clinical trial

A Bayesian methodology is developed to investigate the homogeneity of the treatment effects in a multi-centre clinical trial with an ordinal response. A hierarchical model is formulated for the ordinal response, and the marginal posterior distributions of the covariates, overall treatment and the centre effects are calculated using the Gibbs sampler. The methodology is applied to data arising from a multi-centre clinical trial of therapies for acute myocardial infarction. In this trial, the overall results show that the treatment is effective. However, there appears to be substantial differences in both the baseline risk and treatment effect across centres. Thus, the observed treatment effects may not be generalized to a broader patient population, and exploratory analyses to ascertain reasons for the treatment-by-centre interaction and its possible effect on the study conclusions would be useful.