810 JOURNAL OF THE SOCIETY OF COSMETIC CHEMISTS authors have felt for some time that utilization of modern methods of experimental design and data analysis can be of great value in skin research as well as in practical testing. Surprisingly, very little has been (tone with these techniques in this area, despite the relatively heavy use of statistics in related biological fields.* Many irritants have been patch tested and many investigators have suggested test methods. There are now almost as many methods as there have been investigators. Pioneers have included Schwartz (2), Shelanski (3), Draize (4), Voss (5), and Rostenberg (6), all of whom have made important contributions. However, there remains a need for a standard, easily managed, and statistically valid procedure for the assay of irritation potential, one which will give uniform and reproducible results within a short time. The purposes of this investigation were: 1. To study the advantages of using certain well-known experi- mental designs and subsequent statistical analyses which do not appear to have been previously used in skin testing. 2. To describe certain minor modifications of current human patch testing techniques which the authors feel may improve experi- mental reliability and quantitation. 3. To investigate certain variables which appear to operate in a particular irritant system, principally as an example to demonstrate the experimental techniques and statistical analysis. PRELIMINARY EXPERIMENTS irritant To accomplish the above objectives, particularly the first (viz., to demonstrate the efficiency of formal experimental design and analysis), it was believed advisable to use a specific single irritant. Sodium lauryl sulfate} was chosen because of its common use in many house- hold products and toiletries, because it is believed to have no important sensitization effects, and because it is well-known to dermatological investigators. Scoring Again, in order to use a familiar system, it was felt desirable to have a numerical scale of measurement similar to those used in the past. * Recently, another approach to quantifying skin testing has been described, and will be published soon (16). } Duponol C©--E. I. du Pont de Nemours and Co., Inc., Wilmington, Del. 19898.

STATISTICAL EVALUATION OF CUTANEOUS IRRITANTS 811 It was believed, however, that the degree of irritation could be judged visually much more precisely than has previously been supposed, and the scale shown in Fig. 1 was therefore adopted. In using this scale, the observer was required to estimate the degree of irritation on the treated site compared to untreated surrounding F{oeure I. Irritation scoring scale skin. During the reading of any site, other treated sites were covered with tissue to eliminate direct comparisons. The estimates were always made to the nearest 0.25 unit on the scale of Fig. 1. (It was shown later that this degree of precision was reasonable.) In planning preliminary work in order to identify possible real variables affecting irritation with sodium lauryl sulfate, the following factors were considered' Size of patch Type of covering (occlusion) Nature of patch Quantity of test solution Concentration of irritant Patch contact time Observation time (elapsed time after removal of patch) Tightness of patch Nature of subjects' skin Activity of subjects It was intended that many of the potential variables be held con- stant so that the exposition of the statistical procedure would not be confused because of undue complexity. It is to be understood that in any real system under investigation it may be practical and frequently necessary to test more sources of variation than the number included in this case. The experimental designs then used would simply be exten- sions or modifications of those shown here.* In the present case, some preliminary work was done in order to try to select a small number of important factors from the above list. This work is treated briefly below. Development of Type of Patch Some initial work was done with oval-shaped open patches, such as are used often in routine patch testing. These were applied over • in. * There are many modifications of factorial designs, as well as certain nonfactorial ar- rangements a vailable which do not require extremely large experiments when many variables are to be evaluated.