Some Solutions Inspired by Survey Sampling Theory to Build Effective Clinical Trials

The organisation of a design of experiments, for example, for the realisation of a clinical trial, is crucial. It is often desirable to balance designs so that the means of the covariates are approximately the same in the test and control groups. In survey sampling theory, balanced sampling and calibration are two techniques that improve the precision of estimates. In this paper, we show the links between the two areas. We begin by assessing the gain in precision between a balanced design and a simple random sampling for the least squares estimators and the estimator by differences. We compare rerandomisation techniques and the cube method in order to balance the design. We propose a new method, particularly efficient, which combines the cube method with multivariate matching. A set of simulations is carried out in order to evaluate the different methods. The interest of the calibration is shown even if the design is almost balanced. It is thus shown that tools used by survey statisticians can be useful for experimental designs and clinical trials.