Nuovi indicatori di sbilanciamento e prevedibilità nei disegni sequenziali randomizzati: confronti fra 'biased coin designs' diversi

Abstract

Efron's (1971) Biased Coin Design is a well-known randomization technique of sequential clinical trials for comparing two treatments in the absence of prognostic factors that helps neutralize selection bias while keeping the experiment fairly balanced., Efron's idea has been extended to other types of procedures, but only their asymptotic properties as regards selection bias and imbalance are considered in the literature. In this paper we define new indicators of selection bias and imbalance of the sequential experiment, based on the probability of the "worst possible scenario", which can be computed explicitly for different designs and for any finite sample size n without resorting to simulations. Making use of these measures, we compare the main existing designs with the Adjustable Biased Coin Design (ABCD), recently introduced by two of the present authors. In general, the greater flexibility of the ABCD allows one to choose a sequential procedure that turns out, for all n, to be less predictable and more balanced than the other "biased coin" designs.