Kasahara Toshihiko, Yamamoto Yusuke, Nakashima Natsumi, Imamura Mika, Mizumachi Hideyuki, Suzuki Sho, Aiba Setsuya, Kimura Yutaka, Ashikaga Takao, Kojima Hajime, Ono Atsushi, Matsumoto Kazuhiko
Safety Evaluation Center, Fujifilm Corporation, Minamiashigara, Kanagawa, Japan.
Safety Science Research, Kao Corporation, Ichikai, Tochigi, Japan.
J Appl Toxicol. 2025 Mar;45(3):432-439. doi: 10.1002/jat.4712. Epub 2024 Oct 30.
Most predictive models that use alternatives to animal experiments divide judgements into two classes with a cutoff value for each model. However, if the results of alternative methods are close to the cutoff values, the true result may be ambiguous because of variability in the data. Therefore, the OECD GL497 uses a judgement method that establishes a borderline range (BR) around a cutoff value using a statistical method. However, because there is no detailed description of how the BR is calculated, we clarified two specific points. The scale-constant correction method was used to calculate the median absolute deviation (MAD) around the median. In addition, the bottom-raised transformation method was used when the data were "0" because calculation of the BR requires that all data are logarithmic. Indeed, the BRs for the amino acid derivative reactivity assay (ADRA), interleukin-8 reporter gene assay (IL-8 Luc), and epidermal sensitization assay (EpiSensA) were calculated using data from each validation study. The results showed that the BR for ADRA and IL-8 Luc ranged from 4.1 to 5.9 and 1.25 to 1.57, respectively. Furthermore, the BRs of four genes (ATF3, GCLM, DNAJB4, and IL-8) evaluated using EpiSensA ranged from 10.71 to 21.02, 1.64 to 2.45, 1.61 to 2.52, and 3.11 to 5.16, respectively. The difference (deviation) between the lower and upper BR limits and cutoff value for each alternative method were comparable to those of the alternative methods listed in the guidelines (DPRA, KerarinoSens, and h-CLAT) and thus were considered as adequate.
大多数使用动物实验替代方法的预测模型会将判断分为两类，每个模型都有一个临界值。然而，如果替代方法的结果接近临界值，由于数据的变异性，真实结果可能会模糊不清。因此，经合组织GL497采用了一种判断方法，即使用统计方法在临界值周围建立一个边界范围（BR）。然而，由于没有关于如何计算BR的详细描述，我们阐明了两个具体要点。使用尺度常数校正方法计算中位数周围的中位数绝对偏差（MAD）。此外，当数据为“0”时使用底部提升变换方法，因为BR的计算要求所有数据都是对数形式的。实际上，氨基酸衍生物反应性测定（ADRA）、白细胞介素-8报告基因测定（IL-8 Luc）和表皮致敏测定（EpiSensA）的BR是使用每个验证研究的数据计算得出的。结果表明，ADRA和IL-8 Luc的BR分别为4.1至5.9和1.25至1.57。此外，使用EpiSensA评估的四个基因（ATF3、GCLM、DNAJB4和IL-8）的BR分别为10.71至21.02、1.64至2.45、1.61至2.52和3.11至5.16。每种替代方法的BR下限和上限与临界值之间的差异（偏差）与指南中列出的替代方法（DPRA、KerarinoSens和h-CLAT）相当，因此被认为是合适的。
