ABSTRACT
An adverse event (AE) is any untoward medical occurrence in a subject participating in a clinical study who is administered a medical product. An AE may necessarily relate causally to the treatment. As discussed above, except for the prespecified treatment-class-related AEs, it is usually not appropriate to use significance tests for monitoring safety endpoints due to the issue of multiplicity. The beta-binomial relationship is a convenient model for binary endpoints, because the beta distribution is the conjugate prior of the binomial distribution. Using the posterior probability is a Bayesian approach, which enables us to incorporate expert opinion when making inferences and decisions. Apparently, this study set a stopping rule of 1/20 = 0.05 as compared to the threshold based on the literature for monitoring the study, instead of using the literature to construct a prior distribution. The monitoring design thus failed to consider any variability associated with the observed data.
