Statistical consequences of reducing the number of rabbits utilized in eye irritation testing: data on 67...

Data from 67 six-rabbit eye irritation tests were used to generate two-, three-, and four-rabbit Draize scores. The 15 two-rabbit, 20 three-rabbit, and 15 four-rabbit "subsample" scores for each of the 67 petrochemicals tested were used to establish prediction intervals for the original six-rabbit scores. Prediction interval length shortens with increasing sample size, is widest in the middle portion of the Draize scale, and can be used to select the minimum number of rabbits necessary to satisfy a required level of precision. Also, the ability of each subsample size to correctly classify the test materials according to an in-house irritation classification system was evaluated. Results indicate that subsamples of size 2, 3, 4, and 5 were 88, 93, 95%, and 96% accurate, respectively (compared to six rabbits), at correctly classifying the irritation potential of the materials tested.