Validation Study on Five Cytotoxicity Assays by JSAAE -- II. Statistical Analysis

Takashi Omori1, Kaoru Saijo2, Mayako Kato2, Hiroshi Itagaki3, Makoto Hayashi4, Satoru Miyazaki5, Tadao Ohno2, Hideaki Sugawara5, Noriko Teramoto2, Noriho Tanaka6, Shinobu Wakuri6 and Isao Yoshimura1

1 Department of Management Science, Faculty of Engineering, Science University of Tokyo, 1-3 Kagurazaka, Shinjuku-ku, Tokyo 162;
2 RIKEN Cell Bank, The Institute of Physical and Chemical Research (RIKEN), 3-1-1 Koyadai, Tsukaba Science City, Ibaraki 305;
3 Shiseido Safety & Analytical Research Center,1050 Nippa-cho, Kohoku-Ku, Yokohama-shi 223; 4 Division of Genetics and Mutagenesis, Mitamura Building, 26-1 Kaitaicho, Shinjuku-ku, Tokyo 112-0012 Japan 1-18-1 Kamiyoga, Setagaya-ku, Tokyo 158;
5 Life Science Research Information Laboratory, The Institute of Physical and Chemical Research (RIKEN), 2-1 Hirosawa, Wako-shi, Saitama 351-01;
6 Hatano Research Institute, Food and Drug Safety Center, 729-5 Ochiai, Hadano-shi, Kanagawa 257

Original paper :AATEX 5(1-2):39-58,1998
The validation study organized by JSAAE (JSAAE study) brought to light three main statistical problems, namely, (1) how to obtain a reliable estimate of ED50 from each experiment, (2) how to check the availability of the ED50 estimate obtained from each experiment; and (3) how to evaluate the feasibility of assays included in the JSAAE study.
To resolve the first problem, the authors devised a computer program (LAP-JSAAE) on SAS which incorporated a non-linear least squares method to obtain estimates of ED50 based on a logistic model for raw measurements. To resolve the second problem, the authors set several criteria for checking data which had previously been treated with manual adjustments for trivial errors such as descriptions out of format. They are related to detecting extraordinary large observations, inspecting excessively wide confidence intervals obtained by LAP-JSAAE, checking whether observed responses, for at least one dose, are within the range of 20% and 80%, assessing the lack of fit to the logistic model, and so on. To resolve the last problem, the authors devised the "power-for-distinction" (PFD) which was defined as the ratio of the range of medians to the mean value of the hinge-spreads for log(ED50), where medians and hinge-spreads are from inter-laboratory variation and the range and the mean are from chemical response variation.
After establishing the methods, the authors performed data analysis for the JSAAE study and concluded that the crystal-violet staining assay (CV) with HeLa S3 (SC) cells and the colony formation assay (CF) with HeLa S3 (SC) cells are reliable in the sense that they give high values of the PFD.

Key words: alternatives, chemical irritancy, colony formation assay, crystal-violet staining assay, cytotoxicity assay, Draize eye irritation test, ED5O, inter-laboratory validation, LDH release assay, logistic model, MTT assay, neutral red uptake assay, non-linear least squares method.