Rapid hot-room testing of antiperspirants 401 An obvious solution to the problem is to find'out by experiment the mean weight•of each product used by a large assembly of subjects, and to apply that weight to each experimental axilla. This is not very practicable, since it would have to be done for every product tested. It is also scientifically wrong. The amount of roll-on used in real life is markedly influenced by the topography of the user's axillae, and this can vary very considerably. The mean amount can be grossly excessive for subjects with small, deep axillae, and not nearly enough for subjects with broad, flat axillae. The same is probably true, perhaps to a lesser extent, for aerosols. The application of the mean amount to all axillae is therefore no more realistic than the application of any other constant amount. Another difficulty is that although the ratio of the mean amounts of the two products may be known, it is not justifiable to assume that all users would apply them in that ratio one subject might use a lot of roll-on and a little aerosol while another might do the opposite. The same considerations apply even when the two products are of the same type. Aerosols differ in their non-volatile content, discharge rate, coldness and other charac- teristics, and real users do not use a constant spray time nor a constant discharged weight. The use of either of these in a test method carries the risk of giving an unfair advantage to one or other of the products. The most realistic procedure would be to allow the subjects to apply the products themselves. This carries the obvious risk of increased scatter in the results and a con- sequent need for larger numbers of subjects. On the other hand, any other procedure carries the risk of generating ranking orders different from those experienced in actual use. The method to be described employs a two-second spray applied by a member of the laboratory staff, when aerosol products are being tested. This is justified by the admittedly tenuous arguments that it is what most other laboratories do and it is what the manu- facturers of most aerosol products recommend on the pack. Also from our own crude estimates it seems to be a reasonable approximation to the normal practice of a majority of users. For the infrequent experiments in which products of different types are com- pared, we have adopted the practice of allowing the subjects to apply the products themselves. So far we have too little experience to be able to weigh the advantages and disadvantages of doing so. What must always be borne in mind is that the result found in an individual test is not necessarily the same as that which would have been found had the test been conducted under different conditions. When we say, 'Product A was x•o better than product B', we always imply, 'under the particular conditions of our test'. CHOICE OF PARAMETER TO BE MEASURED The primary measurement is, of course, the number of milligrams of sweat secreted by an axilla during a specified collection period. The parameter of efficacy which this is used to assess may, however, be one of several things. In making a choice it is essential to bear in mind the function of a control group or substrate in any experiment: the function of the control substrate is to provide the best possible estimate of how the test substrate would have responded if it had received the control treatment instead of the test treat- ment. It follows that the two substrates (groups) must be alike with respect to the para- meter to be measured. This means that they must respond in the same way when they both receive no treatment and when they both receive the same treatment. The parameter

402 D. C. Cullum selected must also be as constant as possible, otherwise the random variations which must inevitably occur may be so great as to make it impossible to fulfil these conditions. The parameters to be considered include but are not necessarily limited to the following: sweat weight from an individual axilla sweat weight from an individual subject ratio of sweat weights from right and left axillae of an individual subject ('R/L ratio') mean sweat weight, geometric or arithmetic, for a group of axillae or subjects and mean R/L ratio, geometric or arithmetic, for a group of subjects. The best parameter to use is the one which shows the least variation. In practice the choice may be influenced by other constraints. It is essential to appreciate that none of these parameters is necessarily among those which the user employs to assess efficacy. For example, common sense suggests that in normal use the consumer can observe only inefficacy. That is, if she perceives that she is not sweating, she cannot tell whether her antiperspirant is working or whether she would not have been sweating if she had not used it. If she perceives that she is sweating, however, she is likely to interpret that fact as evidence that her antiperspirant is not working. Secondly, she will observe that she is sweating through some signal such as wet patches on her clothing. The connection between presence or absence of wet patches on clothing and variations in the number of milligrams of sweat on absorbent pads may be a tenuous one. The user's judgement may also be influenced by the various sensory impressions she gains when she applies the product, and by the characteristics and persistence of the perfume. There is therefore no a priori reason for supposing that anti- perspirant efficacy as measured in the hot-room will correlate with efficacy as perceived by the user, because the parameters employed are necessarily different and may be quite unrelated. PROTOCOL The protocol, that is the manner in which the method of measurement is to be applied, is determined by the parameter to be measured, and by any ancillary information which may be required. For example, antiperspirants based on aluminium chlorhydrate often show only a modest effect in tests conducted up to 24 h after the first application (6). With daily or twice-daily application, the effect increases and reaches a plateau after a period which may be as long as 14 days. If it is desired to observe this build-up, repeated hot-room sittings will be necessary, but for routine testing of development products it may be adequate to take only one observation. In this case the result will be influenced by the number and frequency of product applications preceding the test, and a test done after three daily applications will be neither better nor worse than one done after eight twice-daily applications it will merely be different. Cross-over designs have been advocated to minimise the effect of side bias. If this type of design is used it is necessary to allow at least 2 weeks to elapse between the two halves of the test to permit the effect of the first set of treatments to disappear. Provided the number of subjects is large enough, application of each treatment to equal numbers of left and right sides will miniraise side bias to an acceptable extent without the need for a cross-over. Just how many subjects are enough depends not only on the level of statistical discrimination required but also on the need to use a sample reasonably representative of the population at large. It might be possible to devise a procedure which would discriminate significantly between two products whose efficacies differed by only 5•o with the use of say twelve subjects. Twelve subjects, however, would be unlikely to constitute a reliably repre-