Count data obtained in developmental toxicity-teratogenicity studies, such as the number of fetal deaths or the number of malformed fetuses per litter, are routinely analyzed by univariate statistical methods. Such methods ignore the dependence among the different outcomes that are measured in the study. Tests of significance based upon the individual outcomes may lead to incorrect conclusions because of the multiplicity of statistical tests.

A categorical approach which takes into account the dependence structure and an inherent ordering among the outcomes is considered. Because the items (the fetuses) that are classified are sub-sampling units in the study, an ordered categorical procedure cannot be applied directly. Also, the responses among fetuses within the sampling unit (the litter) are not independent. A modification of an ordered categorical approach is used to score the outcomes and derive an index for each sampling unit.

The method used to derive the scores and indices is illustrated. The method of scoring is based upon an approximate solution to the orthogonal decomposition of a cumulative chi-squared statistic. This statistic is calculated for the ordered categorical table. After the scores are obtained, the index for each litter is calculated by taking a weighted average of the scores for the outcomes and the number of fetuses with each observed outcome.

The consideration of several statistical models in which litter size is used as an effect or as a weighting factor provides no support, in the review of actual data, for the theory that litter size affects the severity of the outcome. The assumption of normality is found to be suspect in the review of actual positive data. Further, the simulation study indicates that the nonparametric Kruskal-Wallis test is consistently more powerful than the analysis of variance F-statistic. The results confirm the lack of normality observed in the actual data. These observations lead to the recommendation that nonparametric procedures for the one-way design be used in the analysis of the severity indices.