Selected biostatistical aspects of the validation of in vitro toxicological assays.

An overview is presented on selected biostatistical aspects of the validation of in vitro toxicological assays. Primarily, the statistical analysis of single assays is discussed. Several approaches are compared for the possible non-monotonic dose-response relationship with a priori unknown shapes. The use of confidence intervals instead of p values for toxicologically appropriate decision making is explained. New methods are discussed for demonstrating interlaboratory similarity for dose-response designs are discussed. For validation, the inappropriateness of the concordance coefficient is shown, and sensitive and specificity as well as predictive values are proposed as alternatives. The problem of the missing gold standard is highlighted.