Statistical properties of urn designs in clinical trials

We propose a response-adaptive design, described in terms of urn model, whose allocation proportion converges to prespecified values. The main asymptotic results concerning the urn designs are presented and discussed. We adopt the urn model to implement the random allocation procedure of an experiment that aims at testing the mean effect of two treatments. We conduct a statistical analysis on the inferential performance of different tests and we show that, given a non adaptive test T0, the response adaptive model constructs a test T that is better than T0, in terms of (a) higher power and (b) fewer subjects assigned to the inferior treatment. A retrospective real case study is presented.