A dynamic power prior approach to non-inferiority trials for normal means with unknown variance

Non-inferiority (NI) trials compare new experimental therapies to stan- dard ones (active control). Since historical information on the control treatment is often available, a Bayesian approach to NI trials allows to exploit results from past studies and, eventually, to improve accuracy of inference. Here, we propose the use of a dynamic power prior: the active control treatment’s endpoint is modelled by a power prior distribution, whose informativeness is tuned by a measure of similarity between past and current information. The methodology is evaluated and compared to the frequentist method by simulation; an application to real drug data is available as well.