Statistical planning and analysis considerations in the evaluation of in vitro alternatives to whole animal use for eye irritation testing

Reports from the IRAG Working Groups assessing the current status of in vitro alternatives to whole animal eye irritation tests reflect some common approaches. Although each Working Group studied a particular class of assay, typically all Groups evaluated the in vitro alternatives on the basis of correlation with an in vivo test and the statistical significance of that correlation. However, the data furnished to them by the testing organizations had been obtained with little or no standardization of procedures for testing and evaluation of results. This paper presents issues of design and execution of such test programs that are of statistical concern, including objectives of the evaluation process; limitations of correlation; sources of variation and distinction between actual replication and repeated measurements; evaluation of the predictive ability of in vitro tests; association criteria; and other approaches to such evaluation programs. The distinction between statistical significance and toxicological significance is pointed out. Suggestions are presented for standardization of test protocols, test materials and evaluation programs for both in vitro and in vivo tests. A central repository for such standardization, under the aegis of a government agency, professional standards organization, university or research institute, would be a valuable asset. An addendum is furnished to illustrate assertions about a number of factors influencing correlations and their statistical significance.