NEAR-INFRARED SPECTROSCOPY 201 and clinical evaluations were made at the beginning (TO) and end of the treatment period (T4). Clinical scores. The aspect of the dry skin was evaluated by a trained expert on the basis of the following five criteria: ß "papyrac6" state of the skin ("cigarette paper" aspect) ß roughness (tactile evaluation) ß presence of squames ß presence of scales ("snakeskin" aspect) ß irritation (subclinical inflammation: redness) Each criterion was scored from 0 to 4 according to the degree of severity half points were permitted. The average of the five criteria has been taken as the overall score of each individual. The score for each item was also calculated. Statistical analysis. Correlations between the experimental values were sought by using Spearman's test for nonparametric variables and Pearson's test for parametric variables. Regression lines were constructed from the mean scores (half-point steps), together with the corresponding standard error of the mean (SEM). Three-factors analysis of variance, including a hierarchical analysis [time, product subject (product)], was used to deter- mine the statistical significance of the treatment effects, followed by the Newman-Keul test. The results are presented as means -+ SEM. RESULTS IN VITRO STUDIES The absorbance of the dermis was greater than that of total skin (Figure 2), the differ- ence, i.e., the spectrum of the epidermis, showing the two characteristic water peaks at 1450 and 1936 nm. As expected, the absorbance of the skin increased after removal of the epidermis, partly because of the removal of the least-hydrated layer, the stratum corneum, and partly because of an increase in the volume analyzed due to the greater infrared penetration. Figure 3 shows the correlation (r = 0.98, p 0.001) between the difference in absorbance at 1936 and ! !00 according to the water content of the stratum corneum. The difference in absorbance at the wavelengths used clearly increased with water content, as did the water peak (!936 nm). This difference in absorbance is computed according to the modified Beer-Lambert's law (equation !) and is better than any other one computed using the water peak wavelengths and references. CHARACTERIZATION OF DRY SKIN IN VIVO Figure 4 shows the mean spectra obtained for the study population according to the overall clinical score. Absorbance fell gradually with the increase in the clinical score, and this was particularly marked for the two water peaks (!450 and !936 nm) and less marked in the region between 2000 and 2500 nm. Finally, a multiple correlation study has shown that the absorbance (1936--1100 nm) was better correlated to the dry skin score than the 1450-1100 nm one (r = 0.789

202 JOURNAL OF THE SOCIETY OF COSMETIC CHEMISTS 2.0000 1.8000 1.6000 1.4000 z 1. 2000 0 1.0000 0.8000 0.6000 0.4000 0.2000 •Dermis •- Total Skin i i i i i i i i i i i i i i 1200 1400 1600 1800 2000 2200 2400 WAVELENGTH (nm) Figure 2. NIR absorption spectra of total skin, dermis, and epidermis. versus r = -0.720 for n = 106). The band at 1100 nm is the wavelength at which the absorbance of skin is minimal. We therefore used the difference in absorbance at 1100 and 1936 nm in the rest of the study, which is, as in the in vitro experiment, better than other ones. The regression line in Figure 5, constructed using the overall data obtained at TO (before beginning treatment), is a mathematical representation of the decrease in the spectra illustrated in Figure 4. Despite the large number of values (n = 310), the correlation coefficient (r = -0.536, p 0.001) remained very high. For the overall data at TO, the changes in the overall clinical score (Figure 6) as a function of the individual "scaling" and "roughness" scores correlated strongly with the presence of scales (r = 0.76), but weakly with roughness (r - 0.43). The roughness score no longer changed above an overall score of 3, a phenomenon reflected by the correlations between absorbance and these two individual scores: the "scaling" score was still strongly correlated (r = -0.409) with IR absorbance (Figure 7a), while the "roughness" score was only weakly correlated (Figure 7b) (r = -0.344). It is worth noting that the three other descriptive scores ("papyrac•," squames, irritation) are not correlated either to the difference in absorbance (r = - 0.15, - 0.19, - 0.13) or to the overall score (r = 0.25, 0.20, 0.28). The correlation between electrical conductance and the clinical score was significant but linear only for scores below 2.5 (Figure 8). Beyond the conductance is independent of the score. The correlations between descriptive scores values and conductance concern the "papyrac•" scores (r = -0.331, p 0.01)and roughness (r - -0.370, p 0.01). Figure 9 compares the overall results obtained with the five cosmetic preparations, comparing the treated and untreated legs, for the three following parameters: change in