Assessing non-inferiority for incomplete paired-data under non-ignorable missing mechanism

Testing equivalence of incomplete paired data arises frequently in biomedical studies. Most existing work impose the missing at random assumption, which is not realistic in practice. Two Bayesian approaches for testing the non-inferiority of incomplete paired data under non-ignorable missing mechanism are presented. In addition, Bayesian credible intervals and highest posterior density intervals for the risk difference are constructed. Simulation studies are conducted to evaluate the performance of the two Bayesian testing procedures and the credible intervals. Two datasets are used to illustrate the proposed methods.