Statistical Analysis of Screening Studies in Toxicology with Special Emphasis on Neurotoxicology

Screens have a common set of operating characteristics that are not widely appreciated and that make traditional approaches to statistical analysis both insensitive and inefficient in comparison to other available methods. Traditional methods also do not incorporate additional data as it is generated. Such incorporation would serve to strengthen both the design and analysis processes and is essential in the case of screens. Traditional methods of analysis (contingency tables, rank sum, and ANOVA methods) are overviewed briefly, and their weaknesses are discussed. The concept of power and the factors influencing it are discussed. Alternative approaches to analysis of univariate (control charts and central tendency plots) and multivariate (analog contrast plots and multidimensional cluster plots) data from screens are presented, and their performance is evaluated. The resulting general principles of design and analysis of screens for neurotoxicology are presented. The alternative approaches are shown to be superior to traditional approaches in performance toward meeting the objectives of screens.