Principal components analysis of haematological data from F344 rats with bladder cancer fed N-(ethyl)-all-trans-retinamide.

Several multivariate statistical methods are available which can alleviate the problems of analysing the large volumes of data generated from toxicological experiments. One such technique, principal components analysis, provides a method for exploring the relationships between a number of variables (such as blood parameters) and for eliminating redundant data if strong correlations exist between the characters. It also provides a method for clustering individuals, which may reveal similarities between animals in a treatment group or highlight individual 'outliers'. The application of principal components analysis to a set of haematological data from a trial evaluating the efficacy of a synthetic retinoid against carcinogen-induced bladder cancer in the rat has clearly shown, in two bivariate plots, that while some animals in the carcinogen-treated groups were normal, others were anaemic and that animals fed the synthetic retinoid and killed at 1 year had a microcytic anaemia. A full exploration of the data using conventional univariate statistical analysis would have involved at least 28 graphic representations of the data, as well as the interpretation of more than 130 means and SDs. Principal components analysis provides a valuable additional tool for the statistical analysis and exploration of toxicological data, but it must be used in conjunction with univariate or other multivariate methods if hypothesis testing is required. The use of multivariate techniques in toxicology may best be assessed by their practical application to toxicological data, and this paper presents such an evaluation with the aim of encouraging further exploration of the usefulness of principal components analysis. The raw data on which most analyses have been carried out are given.