The design for statistical analysis of Chernoff/Kavlock developmental toxicity test data was presented. The study design was a completely randomized design with three groups exposed to three different chemicals and a control group. Each exposure group was compared to the control group. Frequency of false significance statements in a large number of comparisons for which each null hypothesis was true, called alpha, was chosen as the comparisonwise error rate. A more liberal selection policy could result in rejection of the null hypothesis when true. Group sample sizes were determined by power calculations based on detecting group differences for testing the hypothesis of equal numbers of live pups per litter. Calculations indicated that if 30 animals/group were used, a difference of two pups/litter could be detected at alpha and beta levels equal to 0.05. Given a low guaranteed pregnancy rate of 50 percent and potential mortality, 50 animals/group was chosen as the initial group size to ensure sufficient numbers of viable litters. Study outcomes were classified as death outcome (toxic, sacrifice, error), pregnancy outcome (pregnant, not pregnant), and litter status (viable, at least one live pup and not viable, no live pups). Resulting combined classifications were: error (accidentally dead); live, not pregnant; dead, pregnant; live, pregnant, non viable litter; and live, pregnant, viable litter. Only pregnant dams who died from toxicity or sacrifice on postnatal day three were used in further statistical analyses. Additional maternal data included weights and length of gestation. Data for pups consisted of number of births, total weight at birth, number of live pups/litter, and total litter weight. Statistical tests were: analysis of variance (ANOVA) for maternal weights; Mann Whitney U- test for distribution of weight gains, length of gestation, comparison of live pups/litter in exposed and control groups, weight of pups, weight gain, and survival of pups from day zero to day three; and Fishers Exact test for cause of death, survival rate, and distribution of viable litters between control and exposed groups.
JeAnne R. Burg, Chief, Statistics Unit, Division of Biomedical and Behavioral Science, National Institute for Occupational Safety and Health, 4676 Columbia Parkway, Cincinnati, OH 45226