Number of animals for sequential testing

Abstract US regulatory agencies have used six animals in eye irritation tests. Analyses of eye irritation tests on pesticides (n = 48), consumer products and cosmetics (n = 53), Marzulli and Ruggles database (n = 139), and cleaning products and ingredients (n = 30) have greatly extended previous investigations of the merit of reducing animal sample size in the eye test. Given the existing scoring system for positive animal responses (corneal opacity ⩾1, iritis ⩾1, conjuctival redness ⩾2 and conjunctival chemosis ⩾2), the accuracy of the classification systems currently used by these agencies was determined. The US Consumer Product Safety Commission, US Food and Drug Administration, and US Occupational Safety and Health Administration use a classification system by which a substance is designated as an irritant when at least four of six animals give a positive response. This decision rule leads to a very high accuracy of at least 99% with essentially no false positive and false negative judgments. In contrast, the system used by the US Environmental Protection Agency pesticide program, in which only one or more of six treated animals result in an irritant decision, has an accuracy of only 50–80% with very high false positive rates. Analyses indicated that test sample size could be reduced to three and still preserve very good accuracy, whereas two-animal and one-animal tests did not give satisfactory responses. A two-stage test, in which two animals are tested and evaluated in the first stage before the need for testing one more animal in the second stage is determined, also demonstrated good operating characteristics. Both the one-stage/three-animal test and the two-stage test deserve consideration.