Considering the Test Performance for Three Class Data Using Linear Discriminant Analysis: A Case Study
Elke Genschow, Andrea Seiler and Horst Spielmann
The results of a European Centre for the Validation of Alternative Methods validation study on the embryonic stem (ES) cell test were used as an example for the test performance for “three-class” data using linear discriminant analysis (LDA). Applying LDA, a biostatistical prediction model was developed to assign test chemicals to three classes of embryotoxicity. Three endpoint values were identified to classify the embryotoxic potential of chemicals: a) the inhibition of differentiation of ES cells into cardiomyocytes (ID50), b) the decrease of viability of adult 3T3 cells (IC503T3), and c) the decrease of viability of ES cells (IC50D3) in an MTT cytotoxicity test. It could be demonstrated that an objective statistical method was required in order to minimise the probability of misclassification. The management team of the validation study defined performance criteria, taking into consideration the hit rate for three classes obtained just by chance. An additional purpose of the study reported here was to calculate a hypothetical prevalence and evaluate its influence on the classification results.