Statistical methodology for screening studies with qualitative/quantitative mixtures.

A statistical method for designing screening studies involving several experimental treatments compared to a standard treatment is developed. The screening study identifies the most promising experimental treatments, which then undergo more rigorous evaluation in a future, larger study. The technique is especially relevant for biopharmaceutical research and development in which phase II clinical trials are conducted to identify the most promising drug regimens, which then move on to phase III of clinical development. It is assumed that the underlying distribution of the primary efficacy random variable is a qualitative/quantitative mixture. The proposed methodology involves calculating the probability of accepting each experimental treatment compared to the standard treatment, where the criterion for acceptance is based on the proportion of qualitative observations and a measure of their location. The probability of acceptance is displayed in two dimensions using operating characteristic contour plots. The techniques are illustrated with some practical examples and some extensions are indicated.