The EpiDerm test protocol for the upcoming ECVAM validation study on in vitro skin irritation tests - An assessment of the performance of the optimised tes

allemand. During the past decade, several validation studies have been conducted on in vitro methods for discriminating between skin irritating and non-irritating chemicals. The reconstructed human skin models, EpiDerm and EPISKIN, provided the most promising results. Based on experience of the similar performance of the two skin models, it was suggested that a common test protocol and prediction model should be developed for the prediction of skin irritation potential with the two models. When the EPISKIN protocol was applied with the EpiDerm model, an acceptable specificity (80%) was achieved, whereas the sensitivity (60%) was low. In 2003, the EPISKIN protocol was further refined by extending the post-incubation period following exposure to test chemicals. This extension and additional technical improvements to the EpiDerm protocol were evaluated with 19 chemicals from the prevalidation study. With the new test design, high sensitivity (80%) and specificity (78%) were obtained. The statistical probability for correct classifications was high, so the test was considered to be ready for formal validation. However, since test optimisation had been conducted with the same test chemicals as were used in the ECVAM prevalidation study, it was decided that the optimisation of the protocol had to be verified with a new set of chemicals. Thus, in the current study, 26 additional chemicals (10 rabbit irritants and 16 non-irritants), which had previously been selected and tested by L'OREAL with EPISKIN, were evaluated in three independent experiments with EpiDerm. With this unbalanced testing set, a specificity of 94%, and a sensitivity of 60% were obtained, while the positive and negative predictivity and accuracy remained almost unchanged (around 80%) in comparison to the in vivo rabbit data. Overall, 45 chemicals (20 irritants and 25 non-irritants) were tested according to the final protocol. The resulting high positive (82%) and negative predictive values (79%) confirmed the reliability (accuracy of 80%) of the improved test protocol of the EpiDerm model