Novel statistical approach for evaluating flow cytometric in vitro micronucleus data

Statistical methods currently recommended for analysis of in vitro micronucleus data are based on small sample sizes. The tests are designed to evaluate linear trends and differences between treated and control samples. When using flow cytometric analysis, >5 times the number of cells are easily evaluated, and the variance estimates from these large samples are small. Application of these recommended tests to large samples resulted in statistically significant outcomes which were not considered to be biologically meaningful. Alternative statistical methods for testing trends and differences among treatments that were either widely used, or sample-size independent, were investigated. Using data from 95 experiments (from 2011-2013) where 19% of the experiments were considered positive, results for the various statistical methods were compared. When using either the recommended or alternate methods, 42-68% of the experiments resulted in statistically significant results (p < 0.05). A new concept was then tested using the same data sets: the "z' factor", designed to identify 'hits' during high throughput screening. Using this simple-to-compute statistic the number of significant calls was reduced to 27%. Then, when combined with a biological criterion based on historical vehicle control data, there was restoration of the original positive frequency (19%). Given the larger sample sizes evaluated using flow cytometry, we have demonstrated that traditional statistical tests may be overly sensitive to small changes in micronucleus induction, and that a simple-to-compute index of separation (z') may be a better tool for analysis, provided that the response is first determined to be biologically meaningful. Environ. Mol. Mutagen. 57:589-604, 2016. © 2016 Wiley Periodicals, Inc.