Supplemental Table 5 : Prediction accuracy using gene expression profiles for classification of drug resistant and sensitive acute lymphoblastic leukemia.

Prediction accuracy in randomly selected cohorts comprising 1/3 of the training set for each antileukemic agent using 30, 50 or 100 reselected top probe sets. The median prediction accuracy is shown with corresponding P-values and the 95 percent confidence interval (C.I.) for prednisolone (PRED), vincristine (VCR), asparaginase (ASP) and daunorubicin (DNR) (a) based on all patients and (b) for only patients with B-lineage acute lymphoblastic leukemia. The prediction accuracies were even higher (c, d), when genes were selected only once from the entire training set, and then used to predict sensitive or resistant phenotypes in randomly selected cohorts comprising 1/3 of the population.